Plusieurs gros ajouts :

- Possibilité de discuter avec un PDF ; RAG ou analyse approfondie. Enlèvement de l'autre outil PDF de discussion qui analysait d'abord un PDF en proposant directement une intégration sans attendre qu'on pose de question
- Mise en place de l'import directement dans les outils dans la sidebar
- Mise en place d'un outil pour créer des tables aléatoires avec possibilité d'utiliser pendant la partie
- Mise en place d'un outil pour mettre en place des PNJ, scènes, chapitre.... directement à partir de la discussion avec le PDF
- Mise en place RAG avec mistal-embeding ou nomic si on utilise ollama
- Mise en place mistral, google en fournisseurs alternatifs pour l'IA dans le cloud
- version 0.11.0-bêta

.gitignore

@@ -108,3 +108,4 @@ docker-compose.override.yml
 # ============================================================================
 relay/
 scripts/bump-version.mjs
+brain/data/notebooks/5.json

brain/app/application/chunking.py

@@ -53,3 +53,27 @@ def _split_oversized(paragraph: str, enc, target_tokens: int) -> list[str]:
     for i in range(0, len(tokens), target_tokens):
         out.append(enc.decode(tokens[i : i + target_tokens]))
     return out
+
+
+def split_in_half(text: str) -> tuple[str, str]:
+    """Coupe `text` en deux moitiés ~égales, de préférence sur un saut de ligne
+    proche du milieu (pour ne pas trancher en plein mot/phrase).
+
+    Sert au repli anti-troncature des imports : quand la SORTIE d'un morceau est
+    coupée (le modèle ne peut pas tout réécrire en une réponse), on retraite ce
+    morceau en deux moitiés. Renvoie ('', '') si le texte est trop court pour
+    être découpé utilement (garde-fou anti-récursion infinie).
+    """
+    text = text.strip()
+    if len(text) < 400:
+        return "", ""
+    mid = len(text) // 2
+    # Cherche un saut de ligne juste avant le milieu, sinon juste après.
+    cut = text.rfind("\n", 0, mid)
+    if cut < len(text) // 4:
+        nxt = text.find("\n", mid)
+        cut = nxt if nxt != -1 else mid
+    left, right = text[:cut].strip(), text[cut:].strip()
+    if not left or not right:
+        return "", ""
+    return left, right

brain/app/application/embeddings.py

@@ -0,0 +1,20 @@
+"""Port d'embeddings (RAG des notebooks).
+
+Abstraction du calcul de vecteurs : un texte → une liste de floats. Les adapters
+concrets (Ollama local, Mistral cloud) la satisfont par duck typing, comme pour
+les LLMProvider. Le RAG n'en dépend que via cette interface.
+"""
+from __future__ import annotations
+
+from typing import Protocol
+
+
+class EmbeddingError(Exception):
+    """Échec du calcul d'embeddings (modèle indisponible, réseau, quota…)."""
+
+
+class EmbeddingProvider(Protocol):
+    """Calcule les vecteurs d'une liste de textes (ordre préservé)."""
+
+    async def embed(self, texts: list[str]) -> list[list[float]]:
+        ...

brain/app/application/import_campaign.py

@@ -12,9 +12,14 @@ from __future__ import annotations
 
 import logging
 
-from app.application.chunking import chunk_text
+from app.application.chunking import chunk_text, split_in_half
-from app.application.llm_json import load_json_object
+from app.application.llm_json import load_json_object, looks_like_truncated_json
 from app.application.llm_retry import generate_with_retry
+from app.application.streaming import with_heartbeat
+
+# Repli anti-troncature : si la sortie d'un morceau est coupée, on le retraite en
+# 2 moitiés. Borné en profondeur (3 niveaux => jusqu'à 8 sous-blocs).
+_MAX_SPLIT_DEPTH = 3
 from app.domain.models import (
     ArcProposal,
     CampaignImportResult,
@@ -22,11 +27,13 @@ from app.domain.models import (
     RoomProposal,
     SceneProposal,
 )
-from app.domain.ports import LLMProvider, PdfTextExtractor
+from app.domain.ports import LLMProvider, LLMProviderError, PdfTextExtractor
 
 logger = logging.getLogger(__name__)
 
-_TEMPERATURE = 0.2
+# Très basse : structuration = recopie/réorganisation fidèle, pas de créativité.
+# Plus la valeur est haute, plus le modèle "brode" (invente du contenu absent).
+_TEMPERATURE = 0.1
 
 # Nom de l'arc unique quand le livre n'est pas découpé en actes/parties.
 _DEFAULT_ARC_NAME = "Aventure principale"
@@ -227,8 +234,30 @@ class ImportCampaignUseCase:
         }
 
         merger = _TreeMerger()
+        skipped = 0
+        last_error: str | None = None
         for i, chunk in enumerate(chunks):
-            merger.add(await self._map_chunk(chunk, index=i, total=total))
+            # RÉSILIENCE : un morceau qui échoue (provider saturé, quota, etc.) est
+            # SAUTÉ — on ne perd pas tout l'import pour autant. On n'abandonne que
+            # si AUCUN morceau ne passe (cf. après la boucle).
+            # HEARTBEAT : keep-alive pendant l'appel LLM pour ne jamais laisser le
+            # flux SSE silencieux (sinon le Core coupe sur timeout d'inactivité).
+            try:
+                arcs_payload: list[dict] | None = None
+                async for kind, payload in with_heartbeat(
+                    self._map_chunk(chunk, index=i, total=total)
+                ):
+                    if kind == "heartbeat":
+                        yield {"type": "heartbeat", "current": i + 1, "total": total}
+                    else:
+                        arcs_payload = payload
+                merger.add(arcs_payload or [])
+            except LLMProviderError as exc:
+                skipped += 1
+                last_error = str(exc)
+                logger.warning("Morceau %s/%s ignoré (échec LLM) : %s", i + 1, total, exc)
+                yield {"type": "chunk_failed", "current": i + 1, "total": total,
+                       "message": str(exc)[:300]}
             arcs, chapters, scenes = merger.counts()
             yield {
                 "type": "progress",
@@ -237,42 +266,74 @@ class ImportCampaignUseCase:
                 "arc_count": arcs,
                 "chapter_count": chapters,
                 "scene_count": scenes,
+                "skipped": skipped,
             }
 
+        if total > 0 and skipped == total:
+            # Tout a échoué : "done" vide serait trompeur → erreur explicite.
+            yield {"type": "error",
+                   "message": "Tous les morceaux ont échoué auprès du fournisseur IA. "
+                              f"Dernier message : {last_error or 'inconnu'}"}
+            return
+
         yield {
             "type": "done",
             "arcs": _serialize_arcs(merger.result()),
             "page_count": doc.page_count,
             "ocr_page_count": doc.ocr_page_count,
+            "skipped": skipped,
         }
 
     # --- MAP : un morceau → sous-arbre ---------------------------------------
 
     async def _map_chunk(self, chunk: str, *, index: int, total: int) -> list[dict]:
+        return await self._extract_arcs(chunk, index=index, total=total, depth=0)
+
+    async def _extract_arcs(
+        self, text: str, *, index: int, total: int, depth: int
+    ) -> list[dict]:
+        """Extrait l'arborescence d'un texte. Si la SORTIE est tronquée, retraite le
+        texte en DEUX moitiés et concatène — le `_TreeMerger` final dédoublonne par
+        nom (un arc/chapitre coupé entre les moitiés est recollé)."""
         prompt = (
             _MAP_SYSTEM.format(default_arc=_DEFAULT_ARC_NAME)
-            + f"\n\n--- EXTRAIT {index + 1}/{total} ---\n{chunk}\n\n"
+            + f"\n\n--- EXTRAIT {index + 1}/{total} ---\n{text}\n\n"
             "Renvoie maintenant le JSON de l'arborescence."
         )
         raw = await generate_with_retry(
             self._llm, prompt, output_format="json", temperature=_TEMPERATURE)
-        return self._parse_arcs(raw, index=index)
+        arcs, truncated = self._parse_arcs(raw, index=index)
+
+        if truncated and depth < _MAX_SPLIT_DEPTH:
+            left, right = split_in_half(text)
+            if left and right:
+                logger.info(
+                    "Morceau %s : sortie tronquée → re-découpage en 2 moitiés (niveau %s).",
+                    index, depth + 1)
+                a = await self._extract_arcs(left, index=index, total=total, depth=depth + 1)
+                b = await self._extract_arcs(right, index=index, total=total, depth=depth + 1)
+                return a + b
+        if truncated:
+            logger.warning(
+                "Morceau %s : sortie tronquée, profondeur max atteinte — partiel conservé.", index)
+        return arcs
 
     @staticmethod
-    def _parse_arcs(raw: str, *, index: int) -> list[dict]:
+    def _parse_arcs(raw: str, *, index: int) -> tuple[list[dict], bool]:
-        """Parse robuste : objet JSON équilibré, ou récupération partielle si tronqué."""
+        """Parse robuste → (arcs, tronqué). `tronqué`=True si récupération partielle."""
         parsed, recovered = load_json_object(raw)
         if parsed is None:
-            logger.warning("Morceau %s : aucun objet JSON exploitable, ignoré.", index)
+            truncated = looks_like_truncated_json(raw)
-            return []
+            if not truncated:
-        if recovered:
+                logger.warning(
-            logger.warning(
+                    "Morceau %s : aucun objet JSON exploitable, ignoré. "
-                "Morceau %s : sortie tronquée — récupération des éléments complets "
+                    "Début de la réponse du modèle : %r",
-                "(envisagez des morceaux plus petits).", index)
+                    index, (raw or "").strip()[:300] or "(réponse VIDE)")
+            return [], truncated
         if isinstance(parsed, dict):
             arcs = parsed.get("arcs", [])
-            return arcs if isinstance(arcs, list) else []
+            return (arcs if isinstance(arcs, list) else []), recovered
-        return []
+        return [], recovered
 
 
 def _serialize_arcs(arcs: list[ArcProposal]) -> list[dict]:

brain/app/application/import_rules.py

@@ -14,16 +14,25 @@ from __future__ import annotations
 
 import logging
 
-from app.application.chunking import CHUNK_TARGET_TOKENS, chunk_text
+from app.application.chunking import CHUNK_TARGET_TOKENS, chunk_text, split_in_half
-from app.application.llm_json import load_json_object
+from app.application.llm_json import load_json_object, looks_like_truncated_json
 from app.application.llm_retry import generate_with_retry
+from app.application.streaming import with_heartbeat
+
+# Repli anti-troncature : si la SORTIE d'un morceau est coupée (le modèle ne peut
+# pas tout réécrire en une réponse), on retraite ce morceau en 2 moitiés. Borné en
+# profondeur pour éviter une récursion infinie (3 niveaux => jusqu'à 8 sous-blocs ;
+# 1-2 niveaux suffisent en pratique, le reste est un garde-fou).
+_MAX_SPLIT_DEPTH = 3
 from app.domain.models import RulesImportResult
 from app.domain.ports import LLMProvider, LLMProviderError, PdfTextExtractor
 
 logger = logging.getLogger(__name__)
 
 # Température basse : tâche de tri/réécriture fidèle, pas de créativité.
-_TEMPERATURE = 0.2
+# Très basse : structuration = recopie/réorganisation fidèle, pas de créativité.
+# Plus la valeur est haute, plus le modèle "brode" (invente du contenu absent).
+_TEMPERATURE = 0.1
 
 # Taxonomie canonique suggérée au modèle pour homogénéiser les titres entre
 # morceaux (sinon "Combat" / "Le combat" / "Règles de combat" se dispersent).
@@ -93,6 +102,24 @@ class _SectionMerger:
         return {title: "\n\n".join(parts) for title, parts in self._merged.items()}
 
 
+def _combine_sections(a: dict[str, str], b: dict[str, str]) -> dict[str, str]:
+    """Fusionne deux dicts de sections (issus des 2 moitiés d'un morceau re-découpé).
+
+    Titres insensibles à la casse : un même titre présent des deux côtés (une section
+    coupée par le re-découpage) voit ses contenus concaténés au lieu d'être écrasés.
+    """
+    out = dict(a)
+    by_lower = {k.lower(): k for k in out}
+    for title, content in b.items():
+        key = by_lower.get(title.lower())
+        if key is not None:
+            out[key] = f"{out[key]}\n\n{content}".strip()
+        else:
+            out[title] = content
+            by_lower[title.lower()] = title
+    return out
+
+
 class ImportRulesUseCase:
     """Transforme un PDF de règles en proposition de sections markdown."""
 
@@ -150,48 +177,103 @@ class ImportRulesUseCase:
         }
 
         merger = _SectionMerger()
+        skipped = 0
+        last_error: str | None = None
         for i, chunk in enumerate(chunks):
-            new_titles = merger.add(await self._map_chunk(chunk, index=i, total=total))
+            # RÉSILIENCE : un morceau qui échoue est SAUTÉ, l'import continue.
+            # Abandon seulement si AUCUN morceau ne passe (cf. après la boucle).
+            # HEARTBEAT : on émet des keep-alive pendant l'appel LLM (long sur un
+            # provider lent) pour que le flux SSE ne soit jamais coupé par le Core.
+            new_titles: list[str] = []
+            try:
+                sections: dict[str, str] | None = None
+                async for kind, payload in with_heartbeat(
+                    self._map_chunk(chunk, index=i, total=total)
+                ):
+                    if kind == "heartbeat":
+                        yield {"type": "heartbeat", "current": i + 1, "total": total}
+                    else:
+                        sections = payload
+                new_titles = merger.add(sections or {})
+            except LLMProviderError as exc:
+                skipped += 1
+                last_error = str(exc)
+                logger.warning("Morceau %s/%s ignoré (échec LLM) : %s", i + 1, total, exc)
+                yield {"type": "chunk_failed", "current": i + 1, "total": total,
+                       "message": str(exc)[:300]}
             yield {
                 "type": "progress",
                 "current": i + 1,
                 "total": total,
                 "new_sections": new_titles,
+                "skipped": skipped,
             }
 
+        if total > 0 and skipped == total:
+            yield {"type": "error",
+                   "message": "Tous les morceaux ont échoué auprès du fournisseur IA. "
+                              f"Dernier message : {last_error or 'inconnu'}"}
+            return
+
         yield {
             "type": "done",
             "sections": merger.result(),
             "page_count": doc.page_count,
             "ocr_page_count": doc.ocr_page_count,
+            "skipped": skipped,
         }
 
     # --- MAP : un morceau → sections -----------------------------------------
 
     async def _map_chunk(self, chunk: str, *, index: int, total: int) -> dict[str, str]:
+        return await self._extract_sections(chunk, index=index, total=total, depth=0)
+
+    async def _extract_sections(
+        self, text: str, *, index: int, total: int, depth: int
+    ) -> dict[str, str]:
+        """Extrait les sections d'un texte. Si la SORTIE est tronquée, retraite le
+        texte en DEUX moitiés (chacune produit une réponse complète) et fusionne —
+        ainsi aucune section n'est perdue, quel que soit le plafond de sortie."""
         prompt = (
             _MAP_SYSTEM.format(
                 canonical="\n".join(f"  - {s}" for s in _CANONICAL_SECTIONS)
             )
-            + f"\n\n--- EXTRAIT {index + 1}/{total} ---\n{chunk}\n\n"
+            + f"\n\n--- EXTRAIT {index + 1}/{total} ---\n{text}\n\n"
             "Renvoie maintenant le JSON des sections."
         )
         raw = await generate_with_retry(
             self._llm, prompt, output_format="json", temperature=_TEMPERATURE)
-        return self._parse_sections(raw, index=index)
+        sections, truncated = self._parse_sections(raw, index=index)
+
+        if truncated and depth < _MAX_SPLIT_DEPTH:
+            left, right = split_in_half(text)
+            if left and right:
+                logger.info(
+                    "Morceau %s : sortie tronquée → re-découpage en 2 moitiés (niveau %s).",
+                    index, depth + 1)
+                a = await self._extract_sections(left, index=index, total=total, depth=depth + 1)
+                b = await self._extract_sections(right, index=index, total=total, depth=depth + 1)
+                return _combine_sections(a, b)
+        if truncated:
+            logger.warning(
+                "Morceau %s : sortie tronquée, profondeur max atteinte — partiel conservé.", index)
+        return sections
 
     @staticmethod
-    def _parse_sections(raw: str, *, index: int) -> dict[str, str]:
+    def _parse_sections(raw: str, *, index: int) -> tuple[dict[str, str], bool]:
-        """Parse robuste : objet JSON équilibré, ou récupération partielle si tronqué."""
+        """Parse robuste → (sections, tronqué). `tronqué`=True si récupération partielle."""
         parsed, recovered = load_json_object(raw)
         if parsed is None:
-            logger.warning("Morceau %s : aucun objet JSON exploitable, ignoré.", index)
+            # Rien d'exploitable : soit prose (échec), soit JSON coupé avant toute
-            return {}
+            # structure complète (→ on signalera 'tronqué' pour re-découper).
-        if recovered:
+            truncated = looks_like_truncated_json(raw)
-            logger.warning(
+            if not truncated:
-                "Morceau %s : sortie tronquée — récupération des sections complètes "
+                logger.warning(
-                "(envisagez des morceaux plus petits).", index)
+                    "Morceau %s : aucun objet JSON exploitable, ignoré. "
+                    "Début de la réponse du modèle : %r",
+                    index, (raw or "").strip()[:300] or "(réponse VIDE)")
+            return {}, truncated
         if not isinstance(parsed, dict):
             logger.warning("Morceau %s : le LLM n'a pas renvoyé un objet, ignoré.", index)
-            return {}
+            return {}, False
-        return {str(k): str(v) for k, v in parsed.items()}
+        return {str(k): str(v) for k, v in parsed.items()}, recovered

brain/app/application/llm_json.py

@@ -13,6 +13,16 @@ et tout ce qui suit. Renvoie None si aucun objet complet n'est trouvé
 from __future__ import annotations
 
 import json
+import re
+
+# Blocs de "réflexion" des modèles raisonneurs (Nemotron, DeepSeek-R1, QwQ…).
+# Leur contenu est de la prose truffée d'accolades qui piège le détecteur de JSON
+# (et n'est jamais la réponse) → on le retire avant toute analyse.
+_REASONING_RE = re.compile(r"<think(?:ing)?>.*?</think(?:ing)?>", re.DOTALL | re.IGNORECASE)
+
+
+def _strip_reasoning(raw: str) -> str:
+    return _REASONING_RE.sub("", raw)
 
 
 def load_json_object(raw: str) -> tuple[object | None, bool]:
@@ -24,6 +34,7 @@ def load_json_object(raw: str) -> tuple[object | None, bool]:
         auquel cas le second élément vaut True.
     (None, False) si rien d'exploitable.
     """
+    raw = _strip_reasoning(raw)
     obj = extract_json_object(raw)
     if obj is not None:
         try:
@@ -39,6 +50,18 @@ def load_json_object(raw: str) -> tuple[object | None, bool]:
     return None, False
 
 
+def looks_like_truncated_json(raw: str) -> bool:
+    """La sortie ressemble-t-elle à un JSON COUPÉ (accolades/crochets non refermés)
+    plutôt qu'à de la prose ? Sert à déclencher un re-découpage même quand RIEN n'a
+    pu être récupéré (cas où le 1er contenu est si long qu'il est coupé avant toute
+    sous-structure complète). On exige un contenu substantiel pour éviter les
+    faux positifs sur une courte réponse non-JSON."""
+    s = (raw or "").strip()
+    if "{" not in s or len(s) < 100:
+        return False
+    return s.count("{") > s.count("}") or s.count("[") > s.count("]")
+
+
 def extract_json_object(raw: str) -> str | None:
     if not raw:
         return None

brain/app/application/llm_retry.py

@@ -12,13 +12,48 @@ from __future__ import annotations
 
 import asyncio
 import logging
+import re
 
 from app.domain.ports import LLMProvider, LLMProviderError
 
 logger = logging.getLogger(__name__)
 
+# 3 tentatives : assez pour absorber un hoquet transitoire, sans s'acharner des
+# minutes sur un modèle durablement lent/saturé (les heartbeats gardent le flux
+# vivant, mais inutile de faire patienter l'utilisateur 15 min pour rien).
 _ATTEMPTS = 3
-_BASE_DELAY_SECONDS = 2.0
+_BASE_DELAY_SECONDS = 3.0
+# Un rate limit (429) "par minute" ne se libère pas en 2-3s : on attend plus
+# longtemps pour ces erreurs-là (le free tier OpenRouter plafonne ~20 req/min).
+_RATE_LIMIT_DELAYS = [10.0, 25.0, 45.0]
+
+
+def _is_rate_limit(exc: LLMProviderError) -> bool:
+    msg = str(exc).lower()
+    return "429" in msg or "rate" in msg or "too many requests" in msg
+
+
+def _is_daily_quota(exc: LLMProviderError) -> bool:
+    """Limite PAR JOUR (vs par minute) : réessayer est inutile, elle ne se libère
+    qu'au reset quotidien. OpenRouter le précise dans le corps du 429."""
+    msg = str(exc).lower()
+    return "per-day" in msg or "per day" in msg or "free-models-per-day" in msg
+
+
+# OpenRouter renvoie souvent le délai conseillé (saturation amont) :
+# "retry_after_seconds": 8  ou  "Retry-After": "8". On le respecte plutôt que
+# d'attendre une durée fixe arbitraire.
+_RETRY_AFTER_RE = re.compile(r'retry[_-]?after(?:_seconds)?"?\s*:\s*"?([0-9]+(?:\.[0-9]+)?)', re.IGNORECASE)
+
+
+def _suggested_retry_after(exc: LLMProviderError) -> float | None:
+    match = _RETRY_AFTER_RE.search(str(exc))
+    if not match:
+        return None
+    try:
+        return float(match.group(1))
+    except ValueError:
+        return None
 
 
 async def generate_with_retry(
@@ -28,7 +63,12 @@ async def generate_with_retry(
     output_format: str | None = None,
     temperature: float | None = None,
 ) -> str:
-    """Comme `llm.generate`, mais réessaie les erreurs transitoires (backoff x2)."""
+    """Comme `llm.generate`, mais réessaie les erreurs transitoires (backoff).
+
+    Backoff plus long pour les 429 (rate limit) afin de laisser la fenêtre se
+    libérer. Nombre de tentatives borné : si le quota est durablement épuisé
+    (ex. limite/jour), l'erreur finit par remonter au lieu de boucler sans fin.
+    """
     delay = _BASE_DELAY_SECONDS
     last_error: LLMProviderError | None = None
     for attempt in range(_ATTEMPTS):
@@ -36,12 +76,27 @@ async def generate_with_retry(
             return await llm.generate(prompt, output_format=output_format, temperature=temperature)
         except LLMProviderError as exc:
             last_error = exc
+            # Quota JOURNALIER épuisé : inutile d'insister, on remonte tout de suite
+            # (sinon on enchaîne des attentes longues pour rien, et on spamme l'API).
+            if _is_daily_quota(exc):
+                logger.warning("Quota journalier du fournisseur épuisé — abandon : %s", exc)
+                raise
             if attempt < _ATTEMPTS - 1:
+                if _is_rate_limit(exc):
+                    suggested = _suggested_retry_after(exc)
+                    if suggested is not None:
+                        # Indication serveur (saturation amont) + petite marge, plafonnée.
+                        wait = min(suggested + 2.0, 60.0)
+                    else:
+                        wait = _RATE_LIMIT_DELAYS[min(attempt, len(_RATE_LIMIT_DELAYS) - 1)]
+                else:
+                    wait = delay
+                    delay *= 2
                 logger.warning(
-                    "Appel LLM échoué (tentative %s/%s) : %s — nouvelle tentative dans %ss.",
+                    "Appel LLM échoué (tentative %s/%s)%s : %s — nouvelle tentative dans %ss.",
-                    attempt + 1, _ATTEMPTS, exc, delay,
+                    attempt + 1, _ATTEMPTS, " [rate limit]" if _is_rate_limit(exc) else "",
+                    exc, wait,
                 )
-                await asyncio.sleep(delay)
+                await asyncio.sleep(wait)
-                delay *= 2
     assert last_error is not None
     raise last_error

brain/app/application/notebook_chat.py

@@ -0,0 +1,90 @@
+"""Use case : chat ANCRÉ sur les sources d'un notebook (RAG).
+
+À chaque message, on retrouve les passages pertinents des sources (via le RAG) et
+on les injecte dans le prompt système, en plus du contexte de campagne. Le modèle
+répond donc en s'appuyant sur la/les source(s) — pas sur ses connaissances générales.
+"""
+from __future__ import annotations
+
+from typing import AsyncIterator
+
+from app.application.notebook_rag import NotebookRagUseCase
+from app.domain.models import ChatMessage
+from app.domain.ports import LLMChatProvider
+
+_SYSTEM_PROMPT = """Tu es un assistant de jeu de rôle qui aide à ADAPTER une source (PDF) à la CAMPAGNE de l'utilisateur.
+
+Tu disposes de DEUX connaissances, toutes deux ci-dessous :
+1) LA CAMPAGNE de l'utilisateur (sa structure arcs/chapitres/scènes, ses PNJ, son univers) ;
+2) LA SOURCE (extraits pertinents du PDF).
+
+Règles :
+- Pour une question sur SA CAMPAGNE (ex. « mon chapitre 3 », « mes PNJ »), appuie-toi sur la section CAMPAGNE.
+- Pour une question sur le livre, appuie-toi sur les EXTRAITS DE LA SOURCE.
+- CROISE les deux pour proposer des adaptations cohérentes avec sa campagne existante.
+- N'invente pas ce qui ne figure ni dans la campagne ni dans la source ; si tu ne sais pas, dis-le.
+- Quand un extrait porte un numéro de page (« (p. 12) »), cite-le (« d'après la p. 12 »).
+
+{context_block}
+--- EXTRAITS PERTINENTS DE LA SOURCE ---
+{sources_block}
+--- FIN DES EXTRAITS ---
+
+PROPOSITIONS D'INTÉGRATION (IMPORTANT) :
+Quand l'utilisateur veut CRÉER ou ADAPTER un élément concret pour sa campagne (un PNJ,
+une scène, un chapitre, un arc, une table aléatoire), termine ta réponse par un ou
+plusieurs BLOCS D'ACTION — un objet JSON par bloc, dans une clôture ```loremind-action.
+L'interface les transformera en boutons « Créer dans la campagne ». N'en mets que si
+c'est pertinent et explicitement souhaité. Formats acceptés :
+
+```loremind-action
+{{"type": "npc", "name": "Nom", "description": "Fiche en quelques phrases."}}
+```
+```loremind-action
+{{"type": "scene", "name": "Nom", "description": "Résumé", "content": "Déroulé détaillé."}}
+```
+```loremind-action
+{{"type": "chapter", "name": "Nom", "description": "Résumé du chapitre."}}
+```
+```loremind-action
+{{"type": "arc", "name": "Nom", "description": "Résumé", "arcType": "LINEAR"}}
+```
+```loremind-action
+{{"type": "table", "name": "Nom", "diceFormula": "1d8", "entries": [{{"minRoll":1,"maxRoll":4,"label":"...","detail":"..."}}]}}
+```
+
+Réponds en français, de façon utile et concise. Mets le texte explicatif AVANT les blocs d'action."""
+
+
+class NotebookChatUseCase:
+    def __init__(self, rag: NotebookRagUseCase, llm: LLMChatProvider) -> None:
+        self._rag = rag
+        self._llm = llm
+
+    async def stream(
+        self,
+        source_ids: list[str],
+        messages: list[ChatMessage],
+        context: str = "",
+        top_k: int = 6,
+    ) -> AsyncIterator[str]:
+        last_user = next((m.content for m in reversed(messages) if m.role == "user"), "")
+        passages = await self._rag.retrieve(source_ids, last_user, top_k=top_k)
+        sources_block = (
+            "\n\n".join(self._format_passage(p) for p in passages)
+            if passages else "(aucun passage pertinent trouvé dans les sources)"
+        )
+        context_block = (
+            f"--- TA CAMPAGNE ---\n{context.strip()}\n--- FIN CAMPAGNE ---\n\n"
+            if context.strip() else "--- TA CAMPAGNE ---\n(aucune donnée de campagne)\n--- FIN CAMPAGNE ---\n\n"
+        )
+        system_prompt = _SYSTEM_PROMPT.format(
+            context_block=context_block, sources_block=sources_block)
+        async for token in self._llm.stream_chat(messages, system_prompt=system_prompt):
+            yield token
+
+    @staticmethod
+    def _format_passage(p: dict) -> str:
+        page = p.get("page")
+        prefix = f"(p. {page}) " if page else ""
+        return f"• {prefix}{p['text'].strip()}"

brain/app/application/notebook_deep.py

@@ -0,0 +1,127 @@
+"""Use case « Analyse approfondie » d'un notebook : map-reduce sur TOUT le document.
+
+Contrairement au chat RAG (qui ne ramène que les top-k extraits), ce mode lit
+l'INTÉGRALITÉ des sources par lots :
+  - MAP    : pour chaque lot, le modèle extrait ce qui est pertinent pour la question
+             (ou « RAS » si rien) ;
+  - REDUCE : il synthétise toutes les notes en une réponse finale (streamée).
+
+→ Répond aux questions globales/exhaustives (« liste tous les… ») quel que soit le
+modèle, au prix de plusieurs appels (comme l'import). Le lot est dimensionné par
+`batch_tokens` (= taille de morceau d'import) : avec un modèle gros-contexte, peu de
+lots ; avec un petit modèle local, plus de lots (mais ça reste exhaustif).
+"""
+from __future__ import annotations
+
+import logging
+from typing import AsyncIterator
+
+import tiktoken
+
+from app.application.llm_retry import generate_with_retry
+from app.domain.models import ChatMessage
+from app.domain.ports import LLMChatProvider, LLMProvider, LLMProviderError
+from app.infrastructure import vector_store
+
+logger = logging.getLogger(__name__)
+
+_NO_MATCH = "RAS"
+_MAP_TEMPERATURE = 0.2
+
+_MAP_PROMPT = """Voici un EXTRAIT d'un document. Extrais UNIQUEMENT les informations
+pertinentes pour répondre à la question ci-dessous. Conserve les détails utiles et
+indique les numéros de page (format « p. X »). Si l'extrait ne contient RIEN de
+pertinent, réponds EXACTEMENT « {no_match} » et rien d'autre.
+
+QUESTION : {question}
+
+--- EXTRAIT ---
+{excerpt}
+--- FIN EXTRAIT ---
+
+Informations pertinentes (ou « {no_match} ») :"""
+
+_REDUCE_SYSTEM = """Tu réponds à la question d'un MJ à partir de NOTES extraites de
+l'ENSEMBLE d'un document source (donc tu as une vue COMPLÈTE, pas un simple extrait).
+Synthétise ces notes en une réponse claire et structurée, cite les pages (« p. X »),
+et n'invente rien qui n'y figure pas. Si une CAMPAGNE est fournie ci-dessous, relie ta
+réponse à sa structure / ses PNJ pour des adaptations cohérentes.
+
+{context_block}
+--- NOTES EXTRAITES DE TOUT LE DOCUMENT ---
+{notes_block}
+--- FIN DES NOTES ---
+
+Réponds en français."""
+
+
+class NotebookDeepUseCase:
+    def __init__(self, llm: LLMProvider, batch_tokens: int = 10000) -> None:
+        self._llm = llm
+        self._batch_tokens = max(2000, batch_tokens)
+
+    async def stream(
+        self,
+        source_ids: list[str],
+        question: str,
+        context: str = "",
+    ) -> AsyncIterator[dict]:
+        """Yield des évènements : {type:'progress',current,total}, {type:'token',token},
+        {type:'done'}. (Les erreurs LLM des lots sont tolérées : lot ignoré.)"""
+        chunks: list[dict] = []
+        for sid in source_ids:
+            chunks.extend(vector_store.all_chunks(sid))
+        if not chunks:
+            yield {"type": "token", "token": "Aucune source indexée à analyser."}
+            yield {"type": "done"}
+            return
+
+        batches = self._group(chunks)
+        total = len(batches)
+        notes: list[str] = []
+        for i, batch in enumerate(batches):
+            yield {"type": "progress", "current": i, "total": total}
+            excerpt = "\n\n".join(
+                f"(p. {c['page']}) {c['text'].strip()}" if c.get("page") else c["text"].strip()
+                for c in batch
+            )
+            prompt = _MAP_PROMPT.format(no_match=_NO_MATCH, question=question, excerpt=excerpt)
+            try:
+                raw = await generate_with_retry(self._llm, prompt, temperature=_MAP_TEMPERATURE)
+            except LLMProviderError as exc:
+                logger.warning("Analyse approfondie : lot %s/%s ignoré : %s", i + 1, total, exc)
+                continue
+            answer = raw.strip()
+            if answer and answer.upper().rstrip(".") != _NO_MATCH:
+                notes.append(answer)
+        yield {"type": "progress", "current": total, "total": total}
+
+        notes_block = "\n\n".join(notes) if notes else "(aucune information pertinente trouvée dans le document)"
+        context_block = (
+            f"--- TA CAMPAGNE (structure, PNJ, univers) ---\n{context.strip()}\n--- FIN CAMPAGNE ---\n\n"
+            if context.strip() else ""
+        )
+        system_prompt = _REDUCE_SYSTEM.format(context_block=context_block, notes_block=notes_block)
+        llm_chat: LLMChatProvider = self._llm  # type: ignore[assignment]
+        async for token in llm_chat.stream_chat(
+            [ChatMessage(role="user", content=question)], system_prompt=system_prompt
+        ):
+            yield {"type": "token", "token": token}
+        yield {"type": "done"}
+
+    def _group(self, chunks: list[dict]) -> list[list[dict]]:
+        """Regroupe les extraits en lots ~`batch_tokens` (compte tiktoken)."""
+        enc = tiktoken.get_encoding("cl100k_base")
+        batches: list[list[dict]] = []
+        current: list[dict] = []
+        current_tokens = 0
+        for c in chunks:
+            t = len(enc.encode(c.get("text", "")))
+            if current and current_tokens + t > self._batch_tokens:
+                batches.append(current)
+                current, current_tokens = [], 0
+            current.append(c)
+            current_tokens += t
+        if current:
+            batches.append(current)
+        return batches

brain/app/application/notebook_rag.py

@@ -0,0 +1,79 @@
+"""Use case RAG des notebooks : indexer une source PDF et retrouver les passages
+pertinents pour une question.
+
+Chaîne d'indexation : PDF → extraction texte (+OCR) → découpage en extraits courts
+→ embeddings → stockage vectoriel (fichier). À la requête : on embed la question
+et on récupère les extraits les plus proches (cosinus) pour ancrer le chat.
+
+Extraits PLUS COURTS que pour l'import (recopie) : ici on veut une granularité fine
+pour que la recherche pointe un passage précis, pas un demi-chapitre.
+"""
+from __future__ import annotations
+
+import logging
+
+from app.application.chunking import chunk_text
+from app.application.embeddings import EmbeddingProvider
+from app.domain.ports import PdfTextExtractor
+from app.infrastructure import vector_store
+
+logger = logging.getLogger(__name__)
+
+_RAG_CHUNK_TOKENS = 600
+# Un extrait avec quasi aucun texte réel (en-tête/pied de page, fragment de numéro
+# de page isolé « 249 250 ») ne sert à rien en RAG → on l'écarte. Seuil bas et
+# conservateur : on ne coupe QUE les fragments quasi-vides, jamais une vraie phrase.
+_MIN_LETTERS = 15
+
+
+def _has_enough_text(piece: str) -> bool:
+    return sum(c.isalpha() for c in piece) >= _MIN_LETTERS
+
+
+class NotebookRagUseCase:
+    def __init__(
+        self,
+        extractor: PdfTextExtractor,
+        embedder: EmbeddingProvider,
+        chunk_target_tokens: int = _RAG_CHUNK_TOKENS,
+    ) -> None:
+        self._extractor = extractor
+        self._embedder = embedder
+        self._chunk_target_tokens = chunk_target_tokens
+
+    async def index_source(self, source_id: str, pdf_bytes: bytes) -> dict:
+        """Extrait, découpe PAR PAGE (pour garder le n° de page → citations), embed
+        et stocke une source. Renvoie un récap."""
+        doc = self._extractor.extract(pdf_bytes)
+        chunks: list[str] = []
+        pages: list[int] = []
+        for page in doc.pages:
+            for piece in chunk_text(page.text, self._chunk_target_tokens):
+                if not _has_enough_text(piece):
+                    continue  # fragment quasi-vide (en-tête/pied/numéro) → ignoré
+                chunks.append(piece)
+                pages.append(page.index + 1)  # n° de page 1-based pour l'affichage
+        logger.info(
+            "Indexation notebook source=%s : %s page(s) (%s OCR), %s extrait(s).",
+            source_id, doc.page_count, doc.ocr_page_count, len(chunks),
+        )
+        if not chunks:
+            vector_store.save(source_id, [], [])
+            return {"chunks": 0, "page_count": doc.page_count, "ocr_page_count": doc.ocr_page_count}
+        vectors = await self._embedder.embed(chunks)
+        count = vector_store.save(source_id, chunks, vectors, pages)
+        return {
+            "chunks": count,
+            "page_count": doc.page_count,
+            "ocr_page_count": doc.ocr_page_count,
+        }
+
+    async def retrieve(self, source_ids: list[str], query: str, top_k: int = 6) -> list[dict]:
+        """Passages les plus pertinents (toutes sources) pour `query`."""
+        ids = [s for s in source_ids if vector_store.exists(s)]
+        if not ids or not query.strip():
+            return []
+        query_vectors = await self._embedder.embed([query])
+        if not query_vectors:
+            return []
+        return vector_store.search(ids, query_vectors[0], top_k)

brain/app/application/streaming.py

@@ -0,0 +1,45 @@
+"""Heartbeats pour garder un flux SSE 'vivant' pendant une coroutine longue.
+
+Problème résolu : pendant un appel LLM lent (import sur provider gratuit), le
+Brain ne produit AUCUN évènement SSE. Le Core (WebClient) ne 'voit aucun item'
+et coupe la connexion sur timeout d'inactivité :
+
+    ReactiveException: Did not observe any item or terminal signal within Nms
+
+C'est le piège classique du SSE long. La parade standard = envoyer un keep-alive
+périodique. `with_heartbeat` exécute une coroutine en émettant un évènement
+'heartbeat' toutes les `interval` secondes tant qu'elle tourne, puis son résultat
+('result', valeur). Le Core remet son chrono à zéro sur n'importe quel évènement
+reçu (même inconnu) → plus de coupure, quelle que soit la lenteur du modèle.
+"""
+from __future__ import annotations
+
+import asyncio
+from typing import Any, AsyncIterator, Awaitable
+
+# Bien sous le timeout d'inactivité du Core (600s) ET de tout proxy (nginx ~60s).
+HEARTBEAT_INTERVAL_SECONDS = 15.0
+
+
+async def with_heartbeat(
+    coro: Awaitable[Any],
+    *,
+    interval: float = HEARTBEAT_INTERVAL_SECONDS,
+) -> AsyncIterator[tuple[str, Any]]:
+    """Exécute `coro` en émettant ('heartbeat', None) toutes les `interval`s tant
+    qu'elle n'est pas terminée, puis ('result', valeur).
+
+    L'exception éventuelle de `coro` est propagée (re-levée par `task.result()`),
+    donc l'appelant peut l'attraper normalement. Si l'itération est abandonnée
+    (client déconnecté), la tâche sous-jacente est annulée.
+    """
+    task: asyncio.Task = asyncio.ensure_future(coro)
+    try:
+        while not task.done():
+            done, _ = await asyncio.wait({task}, timeout=interval)
+            if not done:
+                yield ("heartbeat", None)
+        yield ("result", task.result())
+    finally:
+        if not task.done():
+            task.cancel()

brain/app/core/config.py

@@ -25,8 +25,9 @@ class Settings(BaseSettings):
         extra="ignore",
     )
 
-    # Provider LLM actif. "ollama" = local ; "onemin" = 1min.ai (etage 2).
+    # Provider LLM actif. "ollama" = local ; "onemin" = 1min.ai ;
-    llm_provider: Literal["ollama", "onemin"] = "ollama"
+    # "openrouter" = OpenRouter ; "mistral" = Mistral ; "gemini" = Google Gemini.
+    llm_provider: Literal["ollama", "onemin", "openrouter", "mistral", "gemini"] = "ollama"
 
     ollama_base_url: str = "http://localhost:11434"
     llm_model: str = "gemma4:26b"
@@ -47,6 +48,41 @@ class Settings(BaseSettings):
     onemin_api_key: str = ""
     onemin_model: str = "gpt-4o-mini"
 
+    # OpenRouter (OpenAI-compatible). Cle + modele modifiables depuis l'UI.
+    # Defaut = routeur `openrouter/free` : choisit un modele GRATUIT (0 credit).
+    # Pour un modele precis gratuit : id finissant par `:free`.
+    openrouter_api_key: str = ""
+    openrouter_model: str = "openrouter/free"
+
+    # Mistral (La Plateforme, OpenAI-compatible). Cle + modele modifiables depuis
+    # l'UI. Tier gratuit « Experiment » sur console.mistral.ai (sans CB). Defaut =
+    # mistral-large-latest (128k contexte, bon en francais et en JSON fidele).
+    mistral_api_key: str = ""
+    mistral_model: str = "mistral-large-latest"
+
+    # Google Gemini (endpoint OpenAI-compatible). Cle gratuite sur
+    # aistudio.google.com (sans CB). Defaut = gemini-2.0-flash : ~1M de contexte
+    # (un livre tient en 1-2 appels), rapide, fidele, quota gratuit genereux.
+    gemini_api_key: str = ""
+    gemini_model: str = "gemini-2.0-flash"
+
+    # Embeddings (RAG des notebooks/ateliers). Modele SEPARE du chat.
+    # "ollama" = local (gratuit, illimite, ideal pour indexer un livre = bcp
+    # d'appels) ; "mistral" = cloud EU (mistral-embed, soumis au rate limit).
+    embedding_provider: Literal["ollama", "mistral"] = "ollama"
+    ollama_embedding_model: str = "nomic-embed-text"
+    mistral_embedding_model: str = "mistral-embed"
+    # Au démarrage, si le provider d'embeddings est Ollama et que le modèle n'est
+    # pas présent, le Brain le télécharge automatiquement (en arrière-plan) → le RAG
+    # marche "out of the box" pour un nouvel utilisateur. Désactivable (connexion
+    # limitée, gestion manuelle des modèles).
+    auto_pull_embedding_model: bool = True
+
+    # Nombre d'extraits récupérés par question dans le chat des ateliers (RAG).
+    # Plus haut = plus de couverture pour les questions larges (« liste les… »),
+    # mais prompt plus long. 8 par défaut (montable jusqu'à ~20 sur grand contexte).
+    rag_top_k: int = 8
+
     # Taille cible d'un morceau (en tokens) pour l'import de PDF (regles/campagne).
     # Plus c'est gros, moins il y a de morceaux => moins de fragmentation et un
     # import plus rapide, MAIS il faut que ca tienne dans la fenetre du modele.

brain/app/core/settings_store.py

@@ -29,6 +29,17 @@ _ALLOWED_KEYS = frozenset({
     "llm_num_ctx",
     "onemin_api_key",
     "onemin_model",
+    "openrouter_api_key",
+    "openrouter_model",
+    "mistral_api_key",
+    "mistral_model",
+    "gemini_api_key",
+    "gemini_model",
+    "embedding_provider",
+    "ollama_embedding_model",
+    "mistral_embedding_model",
+    "auto_pull_embedding_model",
+    "rag_top_k",
     "import_chunk_tokens",
 })
 

brain/app/infrastructure/gemini_adapter.py

@@ -0,0 +1,178 @@
+"""Adapter Google Gemini — implémente les ports LLMProvider / LLMChatProvider.
+
+Gemini expose un endpoint COMPATIBLE OpenAI
+(POST {base}/openai/chat/completions, SSE), donc cet adapter est un client
+"OpenAI-compatible" — même structure que les adapters OpenRouter / Mistral.
+
+Tier GRATUIT : clé API sur aistudio.google.com (sans CB). Atout majeur pour
+l'extraction de PDF : un CONTEXTE de ~1M tokens → un livre entier tient en 1-2
+appels, donc quasi aucun morceau perdu et peu de requêtes (limites jamais
+atteintes). Modèle conseillé : `gemini-2.0-flash` (rapide, gros contexte, fidèle).
+"""
+from __future__ import annotations
+
+import asyncio
+import json
+import logging
+from typing import AsyncIterator
+
+import httpx
+
+from app.core.config import Settings
+from app.domain.models import ChatMessage
+from app.domain.ports import LLMProviderError
+
+logger = logging.getLogger(__name__)
+
+_API_URL = "https://generativelanguage.googleapis.com/v1beta/openai/chat/completions"
+
+# Délai max pour le PREMIER token de contenu (échec rapide si le modèle ne produit
+# rien). Gemini répond vite ; 120s est large.
+_FIRST_TOKEN_TIMEOUT_SECONDS = 120.0
+
+
+class GeminiLLMProvider:
+    """Adapter Gemini (OpenAI-compatible) — satisfait LLMProvider et LLMChatProvider."""
+
+    def __init__(self, settings: Settings) -> None:
+        if not settings.gemini_api_key:
+            raise LLMProviderError(
+                "Clé API Gemini manquante. Configure-la depuis l'écran Paramètres "
+                "(clé gratuite sur aistudio.google.com)."
+            )
+        self._api_key = settings.gemini_api_key
+        self._model = settings.gemini_model
+        self._timeout = settings.llm_timeout_seconds
+
+    def _headers(self) -> dict[str, str]:
+        return {
+            "Authorization": f"Bearer {self._api_key}",
+            "Content-Type": "application/json",
+            "Accept": "application/json",
+        }
+
+    async def generate(
+        self,
+        prompt: str,
+        *,
+        output_format: str | None = None,
+        temperature: float | None = None,
+    ) -> str:
+        """One-shot via streaming (puis recollage), avec garde-fous au temps écoulé."""
+        return await self._collect_with_timeouts(
+            [ChatMessage(role="user", content=prompt)], temperature, output_format
+        )
+
+    async def _collect_with_timeouts(
+        self,
+        messages: list[ChatMessage],
+        temperature: float | None,
+        output_format: str | None,
+    ) -> str:
+        """Collecte le stream avec deux garde-fous : 1er token borné (échec rapide
+        si rien ne sort) + ceiling global `self._timeout`."""
+        async def _collect() -> str:
+            chunks: list[str] = []
+            agen = self._stream(messages, None, temperature, output_format)
+            try:
+                while True:
+                    first = _FIRST_TOKEN_TIMEOUT_SECONDS if not chunks else None
+                    try:
+                        token = await asyncio.wait_for(agen.__anext__(), timeout=first)
+                    except StopAsyncIteration:
+                        break
+                    except asyncio.TimeoutError:
+                        raise LLMProviderError(
+                            f"Erreur Gemini : aucun contenu produit en "
+                            f"{int(_FIRST_TOKEN_TIMEOUT_SECONDS)}s. Réessayez ou vérifiez "
+                            "votre quota gratuit."
+                        )
+                    chunks.append(token)
+            finally:
+                await agen.aclose()
+            return "".join(chunks)
+
+        try:
+            return await asyncio.wait_for(_collect(), timeout=self._timeout)
+        except asyncio.TimeoutError as exc:
+            raise LLMProviderError(
+                f"Erreur Gemini : génération non terminée en {self._timeout}s. Réduisez la "
+                "taille des morceaux d'import ou augmentez le timeout."
+            ) from exc
+
+    async def stream_chat(
+        self,
+        messages: list[ChatMessage],
+        *,
+        system_prompt: str | None = None,
+        temperature: float | None = None,
+    ) -> AsyncIterator[str]:
+        async for token in self._stream(messages, system_prompt, temperature):
+            yield token
+
+    async def _stream(
+        self,
+        messages: list[ChatMessage],
+        system_prompt: str | None,
+        temperature: float | None,
+        output_format: str | None = None,
+    ) -> AsyncIterator[str]:
+        payload_messages: list[dict[str, str]] = []
+        if system_prompt:
+            payload_messages.append({"role": "system", "content": system_prompt})
+        for m in messages:
+            payload_messages.append({"role": m.role, "content": m.content})
+
+        body: dict[str, object] = {
+            "model": self._model,
+            "messages": payload_messages,
+            "stream": True,
+        }
+        if temperature is not None:
+            body["temperature"] = temperature
+
+        async with httpx.AsyncClient(timeout=self._timeout) as client:
+            try:
+                async with client.stream(
+                    "POST", _API_URL, headers=self._headers(), json=body
+                ) as response:
+                    if response.status_code >= 400:
+                        detail = (await response.aread()).decode("utf-8", "replace").strip()
+                        raise LLMProviderError(
+                            f"Erreur Gemini (HTTP {response.status_code})"
+                            + (f" : {detail[:500]}" if detail else "")
+                        )
+                    async for token in self._parse_sse(response):
+                        yield token
+            except httpx.HTTPError as exc:
+                raise LLMProviderError(self._format_http_error(exc)) from exc
+
+    @staticmethod
+    async def _parse_sse(response: httpx.Response) -> AsyncIterator[str]:
+        """SSE OpenAI : lignes `data: {json}`, fin sur `data: [DONE]`."""
+        async for line in response.aiter_lines():
+            if not line or not line.startswith("data:"):
+                continue
+            data = line[len("data:"):].strip()
+            if data == "[DONE]":
+                return
+            try:
+                obj = json.loads(data)
+            except json.JSONDecodeError:
+                continue
+            choices = obj.get("choices")
+            if not choices:
+                continue
+            delta = choices[0].get("delta") or {}
+            content = delta.get("content")
+            if content:
+                yield content
+
+    def _format_http_error(self, exc: httpx.HTTPError) -> str:
+        if isinstance(exc, httpx.TimeoutException):
+            return (
+                f"Erreur Gemini : délai dépassé (timeout {self._timeout}s). Le modèle a "
+                "mis trop de temps — réduis la taille des morceaux d'import ou augmente le timeout."
+            )
+        detail = str(exc) or exc.__class__.__name__
+        return f"Erreur Gemini ({exc.__class__.__name__}) : {detail}"

brain/app/infrastructure/mistral_adapter.py

@@ -0,0 +1,188 @@
+"""Adapter Mistral — implémente les ports LLMProvider / LLMChatProvider.
+
+Mistral (La Plateforme) expose l'API OpenAI standard (POST {base}/chat/completions,
+SSE), donc cet adapter est un client "OpenAI-compatible" — même structure que
+l'adapter OpenRouter. Le `generate` one-shot passe par le streaming (puis
+recollage) avec un timeout au temps écoulé pour ne jamais pendre à l'infini.
+
+Tier GRATUIT : compte sur console.mistral.ai (tier « Experiment »), clé API à
+coller dans l'écran Paramètres. Modèles conseillés pour l'extraction : un grand
+contexte fidèle comme `mistral-large-latest` (128k) ou `mistral-small-latest`.
+"""
+from __future__ import annotations
+
+import asyncio
+import json
+import logging
+from typing import AsyncIterator
+
+import httpx
+
+from app.core.config import Settings
+from app.domain.models import ChatMessage
+from app.domain.ports import LLMProviderError
+
+logger = logging.getLogger(__name__)
+
+_API_URL = "https://api.mistral.ai/v1/chat/completions"
+
+# Délai max pour le PREMIER token de contenu (échec rapide si le modèle est en file
+# d'attente et n'envoie que des keep-alive). Généreux car la file d'un tier gratuit
+# peut être longue.
+_FIRST_TOKEN_TIMEOUT_SECONDS = 120.0
+
+
+class MistralLLMProvider:
+    """Adapter Mistral (OpenAI-compatible) — satisfait LLMProvider et LLMChatProvider."""
+
+    def __init__(self, settings: Settings) -> None:
+        if not settings.mistral_api_key:
+            raise LLMProviderError(
+                "Clé API Mistral manquante. Configure-la depuis l'écran Paramètres."
+            )
+        self._api_key = settings.mistral_api_key
+        self._model = settings.mistral_model
+        self._timeout = settings.llm_timeout_seconds
+
+    def _headers(self) -> dict[str, str]:
+        return {
+            "Authorization": f"Bearer {self._api_key}",
+            "Content-Type": "application/json",
+            "Accept": "application/json",
+        }
+
+    async def generate(
+        self,
+        prompt: str,
+        *,
+        output_format: str | None = None,
+        temperature: float | None = None,
+    ) -> str:
+        """One-shot via streaming (puis recollage) pour robustesse sur longues sorties.
+
+        Timeout au TEMPS ÉCOULÉ (asyncio) en plus du timeout réseau d'httpx :
+        si le provider envoyait des keep-alive sans contenu, l'appel pendrait à
+        l'infini. Ici on coupe net après `self._timeout` secondes.
+        """
+        return await self._collect_with_timeouts(
+            [ChatMessage(role="user", content=prompt)], temperature, output_format
+        )
+
+    async def _collect_with_timeouts(
+        self,
+        messages: list[ChatMessage],
+        temperature: float | None,
+        output_format: str | None,
+    ) -> str:
+        """Collecte le stream avec deux garde-fous au temps écoulé : 1er token borné
+        (file d'attente → échec rapide) + ceiling global `self._timeout`."""
+        async def _collect() -> str:
+            chunks: list[str] = []
+            agen = self._stream(messages, None, temperature, output_format)
+            try:
+                while True:
+                    first = _FIRST_TOKEN_TIMEOUT_SECONDS if not chunks else None
+                    try:
+                        token = await asyncio.wait_for(agen.__anext__(), timeout=first)
+                    except StopAsyncIteration:
+                        break
+                    except asyncio.TimeoutError:
+                        raise LLMProviderError(
+                            f"Erreur Mistral : aucun contenu produit en "
+                            f"{int(_FIRST_TOKEN_TIMEOUT_SECONDS)}s — le modèle est probablement "
+                            "en file d'attente (tier gratuit, 2 req/min). Réessayez plus tard ou "
+                            "choisissez un modèle plus disponible."
+                        )
+                    chunks.append(token)
+            finally:
+                await agen.aclose()
+            return "".join(chunks)
+
+        try:
+            return await asyncio.wait_for(_collect(), timeout=self._timeout)
+        except asyncio.TimeoutError as exc:
+            raise LLMProviderError(
+                f"Erreur Mistral : génération non terminée en {self._timeout}s. Réduisez la "
+                "taille des morceaux d'import, augmentez le timeout, ou changez de modèle."
+            ) from exc
+
+    async def stream_chat(
+        self,
+        messages: list[ChatMessage],
+        *,
+        system_prompt: str | None = None,
+        temperature: float | None = None,
+    ) -> AsyncIterator[str]:
+        async for token in self._stream(messages, system_prompt, temperature):
+            yield token
+
+    async def _stream(
+        self,
+        messages: list[ChatMessage],
+        system_prompt: str | None,
+        temperature: float | None,
+        output_format: str | None = None,
+    ) -> AsyncIterator[str]:
+        payload_messages: list[dict[str, str]] = []
+        if system_prompt:
+            payload_messages.append({"role": "system", "content": system_prompt})
+        for m in messages:
+            payload_messages.append({"role": m.role, "content": m.content})
+
+        body: dict[str, object] = {
+            "model": self._model,
+            "messages": payload_messages,
+            "stream": True,
+        }
+        if temperature is not None:
+            body["temperature"] = temperature
+
+        async with httpx.AsyncClient(timeout=self._timeout) as client:
+            try:
+                async with client.stream(
+                    "POST", _API_URL, headers=self._headers(), json=body
+                ) as response:
+                    if response.status_code >= 400:
+                        # En streaming le corps n'est pas lu automatiquement : on le
+                        # lit pour exposer le détail de Mistral (modèle inconnu, clé
+                        # invalide 401, quota 429…), sinon on n'a que le code HTTP nu.
+                        detail = (await response.aread()).decode("utf-8", "replace").strip()
+                        raise LLMProviderError(
+                            f"Erreur Mistral (HTTP {response.status_code})"
+                            + (f" : {detail[:500]}" if detail else "")
+                        )
+                    async for token in self._parse_sse(response):
+                        yield token
+            except httpx.HTTPError as exc:
+                raise LLMProviderError(self._format_http_error(exc)) from exc
+
+    @staticmethod
+    async def _parse_sse(response: httpx.Response) -> AsyncIterator[str]:
+        """SSE OpenAI : lignes `data: {json}`, fin sur `data: [DONE]`."""
+        async for line in response.aiter_lines():
+            if not line or not line.startswith("data:"):
+                continue  # lignes vides ou keep-alive (`: ...`)
+            data = line[len("data:"):].strip()
+            if data == "[DONE]":
+                return
+            try:
+                obj = json.loads(data)
+            except json.JSONDecodeError:
+                continue
+            choices = obj.get("choices")
+            if not choices:
+                continue
+            delta = choices[0].get("delta") or {}
+            content = delta.get("content")
+            if content:
+                yield content
+
+    def _format_http_error(self, exc: httpx.HTTPError) -> str:
+        """Message lisible (timeout, quota 429, clé invalide 401, modèle inconnu…)."""
+        if isinstance(exc, httpx.TimeoutException):
+            return (
+                f"Erreur Mistral : délai dépassé (timeout {self._timeout}s). Le modèle a "
+                "mis trop de temps — réduis la taille des morceaux d'import ou augmente le timeout."
+            )
+        detail = str(exc) or exc.__class__.__name__
+        return f"Erreur Mistral ({exc.__class__.__name__}) : {detail}"

brain/app/infrastructure/mistral_embedding_adapter.py

@@ -0,0 +1,58 @@
+"""Adapter d'embeddings Mistral (cloud, EU) — POST /v1/embeddings.
+
+Soumis au rate limit du tier gratuit : pour indexer un gros document on envoie
+les textes par lots (et l'appelant peut espacer les appels si besoin).
+"""
+from __future__ import annotations
+
+import httpx
+
+from app.application.embeddings import EmbeddingError
+from app.core.config import Settings
+
+_API_URL = "https://api.mistral.ai/v1/embeddings"
+# Lot raisonnable pour ne pas envoyer un payload géant d'un coup.
+_BATCH_SIZE = 64
+
+
+class MistralEmbeddingProvider:
+    """Implémente EmbeddingProvider via l'API Mistral embeddings."""
+
+    def __init__(self, settings: Settings) -> None:
+        if not settings.mistral_api_key:
+            raise EmbeddingError(
+                "Clé API Mistral manquante (requise pour les embeddings Mistral). "
+                "Configure-la dans les Paramètres ou choisis Ollama pour les embeddings."
+            )
+        self._api_key = settings.mistral_api_key
+        self._model = settings.mistral_embedding_model
+        self._timeout = settings.llm_timeout_seconds
+
+    async def embed(self, texts: list[str]) -> list[list[float]]:
+        if not texts:
+            return []
+        out: list[list[float]] = []
+        headers = {
+            "Authorization": f"Bearer {self._api_key}",
+            "Content-Type": "application/json",
+        }
+        async with httpx.AsyncClient(timeout=self._timeout) as client:
+            for start in range(0, len(texts), _BATCH_SIZE):
+                batch = texts[start:start + _BATCH_SIZE]
+                try:
+                    response = await client.post(
+                        _API_URL, headers=headers, json={"model": self._model, "input": batch})
+                    if response.status_code >= 400:
+                        raise EmbeddingError(
+                            f"Mistral embeddings HTTP {response.status_code} : "
+                            f"{response.text.strip()[:300]}")
+                    data = response.json()
+                except httpx.HTTPError as exc:
+                    raise EmbeddingError(f"Erreur Mistral embeddings : {exc}") from exc
+
+                items = data.get("data")
+                if not isinstance(items, list) or len(items) != len(batch):
+                    raise EmbeddingError("Réponse d'embeddings Mistral inattendue (taille incohérente).")
+                for item in items:
+                    out.append([float(x) for x in item.get("embedding", [])])
+        return out

brain/app/infrastructure/ollama_embedding_adapter.py

@@ -0,0 +1,43 @@
+"""Adapter d'embeddings Ollama (local) — endpoint /api/embed.
+
+Gratuit et illimité (tourne sur la machine). Nécessite d'avoir pullé le modèle
+d'embedding (ex. `ollama pull nomic-embed-text`).
+"""
+from __future__ import annotations
+
+import httpx
+
+from app.application.embeddings import EmbeddingError
+from app.core.config import Settings
+
+
+class OllamaEmbeddingProvider:
+    """Implémente EmbeddingProvider via Ollama /api/embed (batch)."""
+
+    def __init__(self, settings: Settings) -> None:
+        self._base_url = settings.ollama_base_url
+        self._model = settings.ollama_embedding_model
+        self._timeout = settings.llm_timeout_seconds
+
+    async def embed(self, texts: list[str]) -> list[list[float]]:
+        if not texts:
+            return []
+        url = f"{self._base_url}/api/embed"
+        payload = {"model": self._model, "input": texts}
+        async with httpx.AsyncClient(timeout=self._timeout) as client:
+            try:
+                response = await client.post(url, json=payload)
+                if response.status_code >= 400:
+                    body = response.text
+                    raise EmbeddingError(
+                        f"Ollama embeddings HTTP {response.status_code} : {body.strip()[:300]}. "
+                        f"Le modèle '{self._model}' est-il installé ? (ollama pull {self._model})"
+                    )
+                data = response.json()
+            except httpx.HTTPError as exc:
+                raise EmbeddingError(f"Erreur Ollama embeddings : {exc}") from exc
+
+        vectors = data.get("embeddings")
+        if not isinstance(vectors, list) or len(vectors) != len(texts):
+            raise EmbeddingError("Réponse d'embeddings Ollama inattendue (taille incohérente).")
+        return [[float(x) for x in v] for v in vectors]

brain/app/infrastructure/openrouter_adapter.py

@@ -0,0 +1,205 @@
+"""Adapter OpenRouter — implémente les ports LLMProvider / LLMChatProvider.
+
+OpenRouter expose l'API OpenAI standard (POST {base}/chat/completions, SSE), donc
+cet adapter est en réalité un client "OpenAI-compatible". Le `generate` one-shot
+passe lui aussi par le streaming (puis recollage) pour éviter les coupures de
+passerelle sur les longues générations (cf. 1min.ai / Cloudflare 524).
+
+Modèles GRATUITS : utiliser un id finissant par `:free` (ex.
+`meta-llama/llama-3.3-70b-instruct:free`) ou le routeur `openrouter/free` (défaut)
+qui choisit automatiquement un modèle gratuit — aucun crédit consommé.
+"""
+from __future__ import annotations
+
+import asyncio
+import json
+import logging
+from typing import AsyncIterator
+
+import httpx
+
+from app.core.config import Settings
+from app.domain.models import ChatMessage
+
+logger = logging.getLogger(__name__)
+from app.domain.ports import LLMProviderError
+
+_API_URL = "https://openrouter.ai/api/v1/chat/completions"
+
+# Délai max pour le PREMIER token de contenu. Un modèle gratuit "en file d'attente"
+# n'envoie que des keep-alive (aucun contenu) → on échoue vite et clairement au lieu
+# de pendre. Généreux (2 min) car la file d'attente d'un tier gratuit peut être longue.
+_FIRST_TOKEN_TIMEOUT_SECONDS = 120.0
+
+
+class OpenRouterLLMProvider:
+    """Adapter OpenRouter (OpenAI-compatible) — satisfait LLMProvider et LLMChatProvider."""
+
+    def __init__(self, settings: Settings) -> None:
+        if not settings.openrouter_api_key:
+            raise LLMProviderError(
+                "Clé API OpenRouter manquante. Configure-la depuis l'écran Paramètres."
+            )
+        self._api_key = settings.openrouter_api_key
+        self._model = settings.openrouter_model
+        self._timeout = settings.llm_timeout_seconds
+
+    def _headers(self) -> dict[str, str]:
+        return {
+            "Authorization": f"Bearer {self._api_key}",
+            "Content-Type": "application/json",
+            # Attribution facultative (classement OpenRouter) — sans impact fonctionnel.
+            "HTTP-Referer": "https://loremind.app",
+            "X-Title": "LoreMind",
+        }
+
+    async def generate(
+        self,
+        prompt: str,
+        *,
+        output_format: str | None = None,
+        temperature: float | None = None,
+    ) -> str:
+        """One-shot via streaming (puis recollage) pour robustesse sur longues sorties.
+
+        Timeout au TEMPS ÉCOULÉ (asyncio) en plus du timeout réseau d'httpx : un
+        modèle gratuit saturé/en file d'attente envoie des keep-alive (`: OPENROUTER
+        PROCESSING`) mais AUCUN contenu → httpx ne déclenche jamais son read-timeout
+        (des octets arrivent) et l'appel pendrait à l'infini. Ici on coupe net après
+        `self._timeout` secondes, quoi qu'il arrive.
+        """
+        return await self._collect_with_timeouts(
+            [ChatMessage(role="user", content=prompt)], temperature, output_format, "OpenRouter"
+        )
+
+    async def _collect_with_timeouts(
+        self,
+        messages: list[ChatMessage],
+        temperature: float | None,
+        output_format: str | None,
+        provider: str,
+    ) -> str:
+        """Collecte le stream avec DEUX garde-fous au temps écoulé :
+        - 1er token borné (`_FIRST_TOKEN_TIMEOUT_SECONDS`) : détecte un modèle bloqué
+          en file d'attente (que des keep-alive, aucun contenu) → échec rapide ;
+        - ceiling global (`self._timeout`) : génération qui ne se termine jamais.
+        Le timeout réseau d'httpx ne suffit pas : des keep-alive font 'arriver des
+        octets' et empêchent son read-timeout de se déclencher.
+        """
+        async def _collect() -> str:
+            chunks: list[str] = []
+            agen = self._stream(messages, None, temperature, output_format)
+            try:
+                while True:
+                    # Borne SEULEMENT l'attente du 1er token (file d'attente) ; ensuite
+                    # on laisse générer (le ceiling global couvre le reste).
+                    first = _FIRST_TOKEN_TIMEOUT_SECONDS if not chunks else None
+                    try:
+                        token = await asyncio.wait_for(agen.__anext__(), timeout=first)
+                    except StopAsyncIteration:
+                        break
+                    except asyncio.TimeoutError:
+                        raise LLMProviderError(
+                            f"Erreur {provider} : aucun contenu produit en "
+                            f"{int(_FIRST_TOKEN_TIMEOUT_SECONDS)}s — le modèle gratuit est "
+                            "probablement en file d'attente / saturé. Réessayez plus tard ou "
+                            "choisissez un autre modèle (1min.ai, ou payant)."
+                        )
+                    chunks.append(token)
+            finally:
+                await agen.aclose()
+            return "".join(chunks)
+
+        try:
+            return await asyncio.wait_for(_collect(), timeout=self._timeout)
+        except asyncio.TimeoutError as exc:
+            raise LLMProviderError(
+                f"Erreur {provider} : génération non terminée en {self._timeout}s. Réduisez la "
+                "taille des morceaux d'import, augmentez le timeout, ou changez de modèle."
+            ) from exc
+
+    async def stream_chat(
+        self,
+        messages: list[ChatMessage],
+        *,
+        system_prompt: str | None = None,
+        temperature: float | None = None,
+    ) -> AsyncIterator[str]:
+        async for token in self._stream(messages, system_prompt, temperature):
+            yield token
+
+    async def _stream(
+        self,
+        messages: list[ChatMessage],
+        system_prompt: str | None,
+        temperature: float | None,
+        output_format: str | None = None,
+    ) -> AsyncIterator[str]:
+        payload_messages: list[dict[str, str]] = []
+        if system_prompt:
+            payload_messages.append({"role": "system", "content": system_prompt})
+        for m in messages:
+            payload_messages.append({"role": m.role, "content": m.content})
+
+        body: dict[str, object] = {
+            "model": self._model,
+            "messages": payload_messages,
+            "stream": True,
+        }
+        if temperature is not None:
+            body["temperature"] = temperature
+        # NB : on n'impose PAS `response_format=json_object`. Beaucoup de modèles/
+        # providers GRATUITS ne le supportent pas et renvoient une réponse VIDE.
+        # On laisse le modèle répondre librement ; l'extraction JSON en aval
+        # (load_json_object + nettoyage du raisonnement) récupère le JSON dans la prose.
+
+        async with httpx.AsyncClient(timeout=self._timeout) as client:
+            try:
+                async with client.stream(
+                    "POST", _API_URL, headers=self._headers(), json=body
+                ) as response:
+                    if response.status_code >= 400:
+                        # En streaming, le corps n'est pas lu automatiquement : on le
+                        # lit pour exposer le détail d'OpenRouter (ex. le 429 précise
+                        # "free-models-per-day" vs "per-minute"), sinon on n'a que le
+                        # code HTTP nu et le diagnostic est impossible.
+                        detail = (await response.aread()).decode("utf-8", "replace").strip()
+                        raise LLMProviderError(
+                            f"Erreur OpenRouter (HTTP {response.status_code})"
+                            + (f" : {detail[:500]}" if detail else "")
+                        )
+                    async for token in self._parse_sse(response):
+                        yield token
+            except httpx.HTTPError as exc:
+                raise LLMProviderError(self._format_http_error(exc)) from exc
+
+    @staticmethod
+    async def _parse_sse(response: httpx.Response) -> AsyncIterator[str]:
+        """SSE OpenAI : lignes `data: {json}`, fin sur `data: [DONE]`."""
+        async for line in response.aiter_lines():
+            if not line or not line.startswith("data:"):
+                continue  # lignes vides ou commentaires keep-alive (`: ...`)
+            data = line[len("data:"):].strip()
+            if data == "[DONE]":
+                return
+            try:
+                obj = json.loads(data)
+            except json.JSONDecodeError:
+                continue
+            choices = obj.get("choices")
+            if not choices:
+                continue
+            delta = choices[0].get("delta") or {}
+            content = delta.get("content")
+            if content:
+                yield content
+
+    def _format_http_error(self, exc: httpx.HTTPError) -> str:
+        """Message lisible (timeout, quota 429, crédits 402, modèle inconnu…)."""
+        if isinstance(exc, httpx.TimeoutException):
+            return (
+                f"Erreur OpenRouter : délai dépassé (timeout {self._timeout}s). Le modèle a "
+                "mis trop de temps — réduis la taille des morceaux d'import ou augmente le timeout."
+            )
+        detail = str(exc) or exc.__class__.__name__
+        return f"Erreur OpenRouter ({exc.__class__.__name__}) : {detail}"

brain/app/infrastructure/vector_store.py

@@ -0,0 +1,109 @@
+"""Stockage vectoriel fichier (RAG des notebooks) — sans dépendance lourde.
+
+Chaque SOURCE est persistée en un fichier JSON sur le volume `data/` du Brain :
+  data/notebooks/{source_id}.json = {"dim": N, "chunks": [{"text":..., "vector":[...]}]}
+
+À l'échelle d'un livre (quelques centaines d'extraits), une recherche cosinus en
+Python pur est instantanée — inutile d'ajouter numpy/pgvector/une base vectorielle.
+"""
+from __future__ import annotations
+
+import json
+import math
+import re
+from pathlib import Path
+
+_STORE_DIR = Path("data/notebooks")
+_SAFE_ID = re.compile(r"[^A-Za-z0-9_-]")
+
+
+def _path(source_id: str) -> Path:
+    safe = _SAFE_ID.sub("_", str(source_id))
+    return _STORE_DIR / f"{safe}.json"
+
+
+def save(
+    source_id: str,
+    chunks: list[str],
+    vectors: list[list[float]],
+    pages: list[int] | None = None,
+) -> int:
+    """Persiste les (chunk, vecteur[, page]) d'une source. Renvoie le nb d'extraits."""
+    if len(chunks) != len(vectors):
+        raise ValueError("chunks et vectors de tailles différentes")
+    if pages is not None and len(pages) != len(chunks):
+        raise ValueError("pages et chunks de tailles différentes")
+    _STORE_DIR.mkdir(parents=True, exist_ok=True)
+    items = []
+    for i, (c, v) in enumerate(zip(chunks, vectors)):
+        item = {"text": c, "vector": v}
+        if pages is not None:
+            item["page"] = pages[i]
+        items.append(item)
+    payload = {"dim": len(vectors[0]) if vectors else 0, "chunks": items}
+    _path(source_id).write_text(json.dumps(payload, ensure_ascii=False), encoding="utf-8")
+    return len(chunks)
+
+
+def exists(source_id: str) -> bool:
+    return _path(source_id).exists()
+
+
+def delete(source_id: str) -> None:
+    _path(source_id).unlink(missing_ok=True)
+
+
+def _load(source_id: str) -> list[dict]:
+    p = _path(source_id)
+    if not p.exists():
+        return []
+    try:
+        data = json.loads(p.read_text(encoding="utf-8"))
+    except (OSError, json.JSONDecodeError):
+        return []
+    return data.get("chunks", []) if isinstance(data, dict) else []
+
+
+def all_chunks(source_id: str) -> list[dict]:
+    """Tous les extraits d'une source (texte + page), sans vecteurs — pour le mode
+    « analyse approfondie » (map-reduce sur tout le document)."""
+    return [{"text": c.get("text", ""), "page": c.get("page")} for c in _load(source_id)]
+
+
+def _cosine(a: list[float], b: list[float]) -> float:
+    if not a or not b or len(a) != len(b):
+        return 0.0
+    dot = 0.0
+    na = 0.0
+    nb = 0.0
+    for x, y in zip(a, b):
+        dot += x * y
+        na += x * x
+        nb += y * y
+    if na == 0.0 or nb == 0.0:
+        return 0.0
+    return dot / (math.sqrt(na) * math.sqrt(nb))
+
+
+def search(
+    source_ids: list[str],
+    query_vector: list[float],
+    top_k: int = 6,
+) -> list[dict]:
+    """Renvoie les `top_k` extraits les plus proches, toutes sources confondues.
+
+    Chaque résultat : {"text": str, "score": float, "source_id": str}.
+    """
+    scored: list[dict] = []
+    for sid in source_ids:
+        for chunk in _load(sid):
+            vector = chunk.get("vector") or []
+            score = _cosine(query_vector, vector)
+            scored.append({
+                "text": chunk.get("text", ""),
+                "score": score,
+                "source_id": sid,
+                "page": chunk.get("page"),
+            })
+    scored.sort(key=lambda c: c["score"], reverse=True)
+    return scored[:top_k]

brain/app/main.py

@@ -4,7 +4,9 @@ Controller volontairement FIN : il valide l'entrée (DTOs Pydantic), délègue
 au domaine via injection de dépendance (ports + use cases), et transforme les
 erreurs du domaine en réponses HTTP. Aucune connaissance d'Ollama ici.
 """
+import asyncio
 import json
+import logging
 from typing import Annotated, AsyncIterator, Literal
 
 import hmac
@@ -14,11 +16,22 @@ from fastapi import Depends, FastAPI, File, Form, HTTPException, Request, Upload
 from fastapi.responses import JSONResponse, StreamingResponse
 from pydantic import BaseModel, Field
 
+import re
+
 from app.application.adapt_campaign import AdaptCampaignUseCase
 from app.application.chat import ChatUseCase
 from app.application.generate_page import GeneratePageUseCase
 from app.application.import_campaign import ImportCampaignUseCase
 from app.application.import_rules import ImportRulesUseCase
+from app.application.llm_json import load_json_object
+from app.application.llm_retry import generate_with_retry
+from app.application.notebook_rag import NotebookRagUseCase
+from app.application.notebook_chat import NotebookChatUseCase
+from app.application.notebook_deep import NotebookDeepUseCase
+from app.application.embeddings import EmbeddingError
+from app.infrastructure import vector_store
+from app.infrastructure.ollama_embedding_adapter import OllamaEmbeddingProvider
+from app.infrastructure.mistral_embedding_adapter import MistralEmbeddingProvider
 from app.core.config import Settings, get_settings
 from app.core.settings_store import save_overrides
 from app.domain.models import (
@@ -45,14 +58,19 @@ from app.domain.models import (
 from app.domain.ports import LLMProvider, LLMProviderError, PdfExtractionError
 from app.infrastructure.ollama_adapter import OllamaLLMProvider
 from app.infrastructure.onemin_adapter import OneMinAiLLMProvider
+from app.infrastructure.openrouter_adapter import OpenRouterLLMProvider
+from app.infrastructure.mistral_adapter import MistralLLMProvider
+from app.infrastructure.gemini_adapter import GeminiLLMProvider
 from app.infrastructure.pdf_extractor import PyMuPdfTextExtractor
 
 app = FastAPI(
     title="LoreMind Brain",
     description="Backend IA pour la génération de contenu narratif.",
-    version="0.10.0-beta",
+    version="0.11.0-beta",
 )
 
+logger = logging.getLogger(__name__)
+
 
 # Encodeur tiktoken partagé — chargé une fois pour éviter le coût de lookup
 # à chaque requête. On utilise cl100k_base (GPT-3.5/4) comme tokenizer
@@ -354,6 +372,12 @@ def get_llm_provider(
     try:
         if settings.llm_provider == "onemin":
             return OneMinAiLLMProvider(settings)
+        if settings.llm_provider == "openrouter":
+            return OpenRouterLLMProvider(settings)
+        if settings.llm_provider == "mistral":
+            return MistralLLMProvider(settings)
+        if settings.llm_provider == "gemini":
+            return GeminiLLMProvider(settings)
         return OllamaLLMProvider(settings)
     except LLMProviderError as exc:
         # Ex : cle 1min.ai manquante. On renvoie du 400 plutot que du 500
@@ -413,6 +437,38 @@ def get_adapt_campaign_use_case(
         llm=llm, extractor=_PDF_EXTRACTOR, max_input_tokens=settings.import_chunk_tokens)
 
 
+def get_embedding_provider(
+    settings: Annotated[Settings, Depends(get_settings)],
+):
+    """Factory de l'adapter d'embeddings (RAG) selon `embedding_provider`."""
+    try:
+        if settings.embedding_provider == "mistral":
+            return MistralEmbeddingProvider(settings)
+        return OllamaEmbeddingProvider(settings)
+    except EmbeddingError as exc:
+        raise HTTPException(status_code=400, detail=str(exc)) from exc
+
+
+def get_notebook_rag_use_case(
+    embedder: Annotated[object, Depends(get_embedding_provider)],
+) -> NotebookRagUseCase:
+    return NotebookRagUseCase(extractor=_PDF_EXTRACTOR, embedder=embedder)  # type: ignore[arg-type]
+
+
+def get_notebook_chat_use_case(
+    llm: Annotated[LLMProvider, Depends(get_llm_provider)],
+    rag: Annotated[NotebookRagUseCase, Depends(get_notebook_rag_use_case)],
+) -> NotebookChatUseCase:
+    return NotebookChatUseCase(rag=rag, llm=llm)  # type: ignore[arg-type]
+
+
+def get_notebook_deep_use_case(
+    llm: Annotated[LLMProvider, Depends(get_llm_provider)],
+    settings: Annotated[Settings, Depends(get_settings)],
+) -> NotebookDeepUseCase:
+    return NotebookDeepUseCase(llm=llm, batch_tokens=settings.import_chunk_tokens)
+
+
 # --- Endpoints ---
 
 
@@ -422,6 +478,54 @@ def health() -> dict[str, str]:
     return {"status": "ok", "service": "brain"}
 
 
+@app.on_event("startup")
+async def _auto_install_embedding_model() -> None:
+    """Au démarrage : si le provider d'embeddings est Ollama et que le modèle n'est
+    pas installé, on le télécharge EN ARRIÈRE-PLAN → le RAG marche d'emblée pour un
+    nouvel utilisateur, sans bloquer le démarrage du Brain. Best-effort (Ollama peut
+    être absent / la connexion limitée) ; désactivable via `auto_pull_embedding_model`.
+    """
+    settings = get_settings()
+    if not settings.auto_pull_embedding_model or settings.embedding_provider != "ollama":
+        return
+    asyncio.create_task(_ensure_ollama_embedding_model(settings.ollama_base_url, settings.ollama_embedding_model))
+
+
+async def _ensure_ollama_embedding_model(base_url: str, model: str) -> None:
+    # Attend qu'Ollama soit joignable (ordre de démarrage des conteneurs), puis
+    # vérifie la présence du modèle avant de le tirer.
+    for attempt in range(10):
+        try:
+            async with httpx.AsyncClient(timeout=10) as client:
+                tags = await client.get(f"{base_url}/api/tags")
+                tags.raise_for_status()
+                names = [m.get("name", "") for m in tags.json().get("models", [])]
+            if any(n == model or n.startswith(model + ":") for n in names):
+                logger.info("Modèle d'embedding '%s' déjà présent.", model)
+                return
+            break  # Ollama joignable, modèle absent → on tire (ci-dessous)
+        except httpx.HTTPError:
+            await asyncio.sleep(min(5 * (attempt + 1), 30))
+    else:
+        logger.warning(
+            "Ollama injoignable au démarrage — modèle d'embedding '%s' non auto-installé "
+            "(il sera tirable manuellement : ollama pull %s).", model, model)
+        return
+
+    logger.info("Téléchargement automatique du modèle d'embedding '%s'…", model)
+    try:
+        async with httpx.AsyncClient(timeout=None) as client:
+            async with client.stream("POST", f"{base_url}/api/pull", json={"name": model}) as resp:
+                resp.raise_for_status()
+                async for _line in resp.aiter_lines():
+                    pass  # on draine la progression NDJSON jusqu'à la fin
+        logger.info("Modèle d'embedding '%s' prêt.", model)
+    except httpx.HTTPError as exc:
+        logger.warning(
+            "Auto-installation du modèle d'embedding '%s' échouée : %s "
+            "(tirage manuel possible : ollama pull %s).", model, exc, model)
+
+
 @app.post("/generate", response_model=GenerateResponse)
 async def generate(
     body: GenerateRequest,
@@ -552,6 +656,11 @@ async def import_rules_stream(
             yield _sse("error", {"message": str(exc)})
         except LLMProviderError as exc:
             yield _sse("error", {"message": str(exc)})
+        except Exception as exc:  # noqa: BLE001 — filet : une erreur inattendue ne doit
+            # PAS casser le flux SSE brutalement (sinon le Core n'a qu'un message générique
+            # sans détail). On la transforme en évènement `error` propre + log avec trace.
+            logger.exception("Import règles : erreur inattendue dans le flux.")
+            yield _sse("error", {"message": f"Erreur inattendue du Brain : {type(exc).__name__} : {exc}"})
 
     return StreamingResponse(event_stream(), media_type="text/event-stream")
 
@@ -587,6 +696,10 @@ async def import_campaign_stream(
             yield _sse("error", {"message": str(exc)})
         except LLMProviderError as exc:
             yield _sse("error", {"message": str(exc)})
+        except Exception as exc:  # noqa: BLE001 — voir import règles : on ne laisse pas
+            # une erreur inattendue casser le flux sans détail.
+            logger.exception("Import campagne : erreur inattendue dans le flux.")
+            yield _sse("error", {"message": f"Erreur inattendue du Brain : {type(exc).__name__} : {exc}"})
 
     return StreamingResponse(event_stream(), media_type="text/event-stream")
 
@@ -688,7 +801,9 @@ async def chat_stream(
         "system": _count_tokens(system_prompt_preview),
         "history": sum(_count_tokens(m.content) for m in history_msgs),
         "current": _count_tokens(current_msg.content) if current_msg else 0,
-        "max": settings.llm_num_ctx,
+        # Plafond connu seulement pour Ollama (num_ctx). Pour le cloud (1min/OpenRouter)
+        # on ne connaît pas la fenêtre réelle → 0 = "pas de max" (jauge sans dénominateur).
+        "max": settings.llm_num_ctx if settings.llm_provider == "ollama" else 0,
     }
 
     async def event_stream() -> AsyncIterator[str]:
@@ -768,6 +883,245 @@ async def summarize_conversation_title(
     return SummarizeTitleResponseDTO(title=title)
 
 
+# --- Tables aléatoires : génération IA + improvisation -----------------------
+
+_DICE_FORMULA_RE = re.compile(r"^\s*(\d*)\s*[dD]\s*(\d+)\s*$")
+
+
+def _dice_total_range(formula: str) -> tuple[int, int] | None:
+    """(min, max) des totaux possibles d'une formule NdM, ou None si invalide."""
+    match = _DICE_FORMULA_RE.match(formula or "")
+    if not match:
+        return None
+    count = int(match.group(1)) if match.group(1) else 1
+    faces = int(match.group(2))
+    if count < 1 or count > 100 or faces < 2 or faces > 10000:
+        return None
+    return count, count * faces
+
+
+class GenerateTableRequestDTO(BaseModel):
+    description: str
+    dice_formula: str = Field(default="1d20")
+    # Contexte libre assemblé par le Core (nom de campagne, système, ambiance…).
+    context: str = Field(default="")
+
+
+class GeneratedTableEntryDTO(BaseModel):
+    min_roll: int
+    max_roll: int
+    label: str
+    detail: str = ""
+
+
+class GenerateTableResponseDTO(BaseModel):
+    name: str
+    description: str = ""
+    entries: list[GeneratedTableEntryDTO]
+
+
+@app.post("/generate/random-table", response_model=GenerateTableResponseDTO)
+async def generate_random_table(
+    body: GenerateTableRequestDTO,
+    llm: Annotated[LLMProvider, Depends(get_llm_provider)],
+) -> GenerateTableResponseDTO:
+    """Génère une table aléatoire (entrées par plage) couvrant la formule de dé."""
+    rng = _dice_total_range(body.dice_formula)
+    if rng is None:
+        raise HTTPException(status_code=422, detail="Formule de dé invalide (ex. 1d20, 2d6, d100).")
+    lo, hi = rng
+    context_block = f"\nContexte de la campagne :\n{body.context.strip()}\n" if body.context.strip() else ""
+    prompt = (
+        "Tu es un assistant de jeu de rôle. Génère une TABLE ALÉATOIRE évocatrice.\n"
+        f"Dé : {body.dice_formula} (résultats possibles de {lo} à {hi}).\n"
+        f"Sujet : {body.description.strip()}\n"
+        f"{context_block}\n"
+        "Règles IMPÉRATIVES :\n"
+        "- Réponds UNIQUEMENT par un objet JSON valide, sans texte autour.\n"
+        '- Format : {"name": "...", "description": "...", "entries": '
+        '[{"min_roll": N, "max_roll": M, "label": "résultat court", "detail": "1-2 phrases"}]}\n'
+        f"- Les plages (min_roll..max_roll) doivent COUVRIR EXACTEMENT {lo}..{hi}, "
+        "sans trou ni chevauchement, dans l'ordre croissant.\n"
+        "- Des résultats variés, cohérents avec le sujet (et le contexte s'il est fourni).\n"
+        "- En français. 'label' = résultat bref ; 'detail' = description/effet concret.\n"
+        "Renvoie maintenant le JSON."
+    )
+    try:
+        raw = await generate_with_retry(llm, prompt, output_format="json", temperature=0.7)
+    except LLMProviderError as exc:
+        raise HTTPException(status_code=502, detail=str(exc)) from exc
+
+    parsed, _ = load_json_object(raw)
+    if not isinstance(parsed, dict):
+        raise HTTPException(status_code=502, detail="Le modèle n'a pas renvoyé de table exploitable.")
+
+    entries: list[GeneratedTableEntryDTO] = []
+    for e in parsed.get("entries", []) or []:
+        if not isinstance(e, dict):
+            continue
+        try:
+            mn = int(e["min_roll"])
+            mx = int(e["max_roll"])
+        except (KeyError, TypeError, ValueError):
+            continue
+        label = str(e.get("label") or "").strip()
+        if not label:
+            continue
+        entries.append(GeneratedTableEntryDTO(
+            min_roll=mn, max_roll=max(mn, mx), label=label[:200],
+            detail=str(e.get("detail") or "").strip(),
+        ))
+    if not entries:
+        raise HTTPException(status_code=502, detail="Aucune entrée générée — réessaie ou reformule.")
+
+    name = str(parsed.get("name") or body.description).strip()[:120] or "Table générée"
+    return GenerateTableResponseDTO(
+        name=name,
+        description=str(parsed.get("description") or "").strip(),
+        entries=entries,
+    )
+
+
+class ImproviseRollRequestDTO(BaseModel):
+    table_name: str
+    result_label: str
+    result_detail: str = Field(default="")
+    context: str = Field(default="")
+
+
+class ImproviseRollResponseDTO(BaseModel):
+    narration: str
+
+
+@app.post("/improvise/table-roll", response_model=ImproviseRollResponseDTO)
+async def improvise_table_roll(
+    body: ImproviseRollRequestDTO,
+    llm: Annotated[LLMProvider, Depends(get_llm_provider)],
+) -> ImproviseRollResponseDTO:
+    """Brode un court récit (2-3 phrases) sur un résultat tiré, pour lancer la scène."""
+    detail = f" ({body.result_detail.strip()})" if body.result_detail.strip() else ""
+    context_block = f"\nContexte : {body.context.strip()}" if body.context.strip() else ""
+    prompt = (
+        "Tu es le Maître du Jeu. Les joueurs viennent de tirer sur la table "
+        f"« {body.table_name.strip()} » et ont obtenu : « {body.result_label.strip()} »{detail}."
+        f"{context_block}\n\n"
+        "Décris en 2-3 phrases vivantes et immédiates ce qui se passe, pour lancer la scène. "
+        "Pas de méta, pas d'options : juste la narration, en français."
+    )
+    try:
+        raw = await llm.generate(prompt, temperature=0.8)
+    except LLMProviderError as exc:
+        raise HTTPException(status_code=502, detail=str(exc)) from exc
+    return ImproviseRollResponseDTO(narration=raw.strip())
+
+
+# --- Notebooks (atelier RAG) : indexation des sources + chat ancré ----------
+
+
+class IndexSourceResponseDTO(BaseModel):
+    chunks: int
+    page_count: int
+    ocr_page_count: int
+
+
+@app.post("/index/notebook-source", response_model=IndexSourceResponseDTO)
+async def index_notebook_source(
+    rag: Annotated[NotebookRagUseCase, Depends(get_notebook_rag_use_case)],
+    source_id: str = Form(...),
+    file: UploadFile = File(...),
+) -> IndexSourceResponseDTO:
+    """Indexe une source PDF (extraction + embeddings + stockage vectoriel)."""
+    content = await file.read()
+    if not content:
+        raise HTTPException(status_code=422, detail="Fichier PDF vide.")
+    if len(content) > _MAX_PDF_BYTES:
+        raise HTTPException(
+            status_code=413, detail=f"PDF trop volumineux (> {_MAX_PDF_BYTES // (1024 * 1024)} Mo).")
+    try:
+        recap = await rag.index_source(source_id, content)
+    except PdfExtractionError as exc:
+        raise HTTPException(status_code=422, detail=str(exc)) from exc
+    except EmbeddingError as exc:
+        raise HTTPException(status_code=502, detail=str(exc)) from exc
+    return IndexSourceResponseDTO(**recap)
+
+
+@app.delete("/index/notebook-source/{source_id}")
+def delete_notebook_source(source_id: str) -> dict[str, str]:
+    """Supprime les vecteurs d'une source (au DELETE d'une source/notebook)."""
+    vector_store.delete(source_id)
+    return {"status": "deleted", "source_id": source_id}
+
+
+class NotebookChatMessageDTO(BaseModel):
+    role: str
+    content: str
+
+
+class NotebookChatRequestDTO(BaseModel):
+    source_ids: list[str] = Field(default_factory=list)
+    messages: list[NotebookChatMessageDTO] = Field(default_factory=list)
+    context: str = Field(default="")
+
+
+@app.post("/chat/notebook/stream")
+async def chat_notebook_stream(
+    body: NotebookChatRequestDTO,
+    use_case: Annotated[NotebookChatUseCase, Depends(get_notebook_chat_use_case)],
+    settings: Annotated[Settings, Depends(get_settings)],
+) -> StreamingResponse:
+    """Chat ANCRÉ sur les sources (RAG) : récupère les passages pertinents puis
+    streame la réponse. Évènements SSE : `token` {token}, `done` {}, `error` {message}."""
+    messages = [ChatMessage(role=m.role, content=m.content) for m in body.messages]
+    top_k = max(1, min(settings.rag_top_k, 200))
+
+    def _sse(event: str, data: dict) -> str:
+        return f"event: {event}\ndata: {json.dumps(data, ensure_ascii=False)}\n\n"
+
+    async def event_stream() -> AsyncIterator[str]:
+        try:
+            async for token in use_case.stream(body.source_ids, messages, context=body.context, top_k=top_k):
+                if token:
+                    yield _sse("token", {"token": token})
+            yield _sse("done", {})
+        except (LLMProviderError, EmbeddingError) as exc:
+            yield _sse("error", {"message": str(exc)})
+        except Exception as exc:  # noqa: BLE001 — filet : pas de coupure brutale du flux.
+            logger.exception("Chat notebook : erreur inattendue.")
+            yield _sse("error", {"message": f"Erreur inattendue du Brain : {type(exc).__name__} : {exc}"})
+
+    return StreamingResponse(event_stream(), media_type="text/event-stream")
+
+
+@app.post("/chat/notebook/deep/stream")
+async def chat_notebook_deep_stream(
+    body: NotebookChatRequestDTO,
+    use_case: Annotated[NotebookDeepUseCase, Depends(get_notebook_deep_use_case)],
+) -> StreamingResponse:
+    """Analyse APPROFONDIE (map-reduce sur tout le document). Évènements SSE :
+    `progress` {current,total} pendant la lecture, puis `token` {token}, puis `done`."""
+    question = next((m.content for m in reversed(body.messages) if m.role == "user"), "")
+
+    def _sse(event: str, data: dict) -> str:
+        return f"event: {event}\ndata: {json.dumps(data, ensure_ascii=False)}\n\n"
+
+    async def event_stream() -> AsyncIterator[str]:
+        if not question.strip():
+            yield _sse("error", {"message": "Question vide."})
+            return
+        try:
+            async for ev in use_case.stream(body.source_ids, question, context=body.context):
+                ev_type = ev.pop("type")
+                yield _sse(ev_type, ev)
+        except (LLMProviderError, EmbeddingError) as exc:
+            yield _sse("error", {"message": str(exc)})
+        except Exception as exc:  # noqa: BLE001 — filet : pas de coupure brutale.
+            logger.exception("Analyse approfondie : erreur inattendue.")
+            yield _sse("error", {"message": f"Erreur inattendue du Brain : {type(exc).__name__} : {exc}"})
+
+    return StreamingResponse(event_stream(), media_type="text/event-stream")
+
+
 # --- Mapping DTO → domaine (frontière HTTP) ---------------------------------
 
 
@@ -885,12 +1239,27 @@ class SettingsDTO(BaseModel):
     Les secrets (onemin_api_key) sont masques en lecture.
     """
 
-    llm_provider: Literal["ollama", "onemin"]
+    llm_provider: Literal["ollama", "onemin", "openrouter", "mistral", "gemini"]
     ollama_base_url: str
     llm_model: str
     onemin_model: str
     # True si une cle 1min.ai est deja configuree — pas de leak de la cle elle-meme.
     onemin_api_key_set: bool
+    openrouter_model: str
+    # True si une cle OpenRouter est deja configuree (cle elle-meme jamais renvoyee).
+    openrouter_api_key_set: bool
+    mistral_model: str
+    # True si une cle Mistral est deja configuree (cle elle-meme jamais renvoyee).
+    mistral_api_key_set: bool
+    gemini_model: str
+    # True si une cle Gemini est deja configuree (cle elle-meme jamais renvoyee).
+    gemini_api_key_set: bool
+    # Embeddings (RAG des ateliers) : provider + modeles + auto-pull Ollama.
+    embedding_provider: Literal["ollama", "mistral"]
+    ollama_embedding_model: str
+    mistral_embedding_model: str
+    auto_pull_embedding_model: bool
+    rag_top_k: int
     # Fenetre de contexte effective passee au modele (num_ctx Ollama) — sert
     # aussi de plafond a la jauge de contexte UI.
     llm_num_ctx: int
@@ -903,12 +1272,23 @@ class SettingsDTO(BaseModel):
 class SettingsUpdateDTO(BaseModel):
     """Patch partiel des settings. Tous les champs sont optionnels."""
 
-    llm_provider: Literal["ollama", "onemin"] | None = None
+    llm_provider: Literal["ollama", "onemin", "openrouter", "mistral", "gemini"] | None = None
     ollama_base_url: str | None = None
     llm_model: str | None = None
     onemin_model: str | None = None
     # Chaine vide => on efface la cle. None => pas de changement.
     onemin_api_key: str | None = None
+    openrouter_model: str | None = None
+    openrouter_api_key: str | None = None
+    mistral_model: str | None = None
+    mistral_api_key: str | None = None
+    gemini_model: str | None = None
+    gemini_api_key: str | None = None
+    embedding_provider: Literal["ollama", "mistral"] | None = None
+    ollama_embedding_model: str | None = None
+    mistral_embedding_model: str | None = None
+    auto_pull_embedding_model: bool | None = None
+    rag_top_k: int | None = None
     llm_num_ctx: int | None = None
     import_chunk_tokens: int | None = None
     llm_timeout_seconds: int | None = None
@@ -921,6 +1301,17 @@ def _to_settings_dto(s: Settings) -> SettingsDTO:
         llm_model=s.llm_model,
         onemin_model=s.onemin_model,
         onemin_api_key_set=bool(s.onemin_api_key),
+        openrouter_model=s.openrouter_model,
+        openrouter_api_key_set=bool(s.openrouter_api_key),
+        mistral_model=s.mistral_model,
+        mistral_api_key_set=bool(s.mistral_api_key),
+        gemini_model=s.gemini_model,
+        gemini_api_key_set=bool(s.gemini_api_key),
+        embedding_provider=s.embedding_provider,
+        ollama_embedding_model=s.ollama_embedding_model,
+        mistral_embedding_model=s.mistral_embedding_model,
+        auto_pull_embedding_model=s.auto_pull_embedding_model,
+        rag_top_k=s.rag_top_k,
         llm_num_ctx=s.llm_num_ctx,
         import_chunk_tokens=s.import_chunk_tokens,
         llm_timeout_seconds=s.llm_timeout_seconds,
@@ -1081,6 +1472,144 @@ async def delete_ollama_model(
     return {"status": "deleted", "name": name}
 
 
+@app.get("/models/openrouter")
+async def list_openrouter_models() -> dict[str, list[dict[str, object]]]:
+    """Catalogue DYNAMIQUE des modeles OpenRouter (API publique, sans cle).
+
+    Renvoie {models: [{id, name, context_length, free}]}, trie gratuits d'abord
+    puis contexte decroissant. `free` = id finissant par ':free' OU prix nul.
+    """
+    try:
+        async with httpx.AsyncClient(timeout=20) as client:
+            response = await client.get("https://openrouter.ai/api/v1/models")
+            response.raise_for_status()
+            data = response.json()
+    except httpx.HTTPError as exc:
+        raise HTTPException(status_code=502, detail=f"OpenRouter injoignable : {exc}")
+
+    def _is_zero(value: object) -> bool:
+        try:
+            return float(value) == 0.0  # type: ignore[arg-type]
+        except (TypeError, ValueError):
+            return False
+
+    models: list[dict[str, object]] = []
+    for m in data.get("data", []) or []:
+        mid = str(m.get("id") or "")
+        if not mid:
+            continue
+        pricing = m.get("pricing") or {}
+        is_free = mid.endswith(":free") or (
+            _is_zero(pricing.get("prompt")) and _is_zero(pricing.get("completion"))
+        )
+        try:
+            ctx = int(m.get("context_length") or 0)
+        except (TypeError, ValueError):
+            ctx = 0
+        models.append({
+            "id": mid,
+            "name": str(m.get("name") or mid),
+            "context_length": ctx,
+            "free": is_free,
+        })
+
+    models.sort(key=lambda x: (not x["free"], -int(x["context_length"])))  # type: ignore[index]
+    return {"models": models}
+
+
+# Repli statique si la cle Mistral n'est pas (encore) configuree ou si l'API est
+# injoignable — l'utilisateur peut quand meme choisir un modele. Liste curee
+# (juin 2026) ; pour l'extraction de PDF, prefere `large` (fidele, 128k) ou `small`.
+_MISTRAL_FALLBACK_MODELS = [
+    "mistral-large-latest",
+    "mistral-medium-latest",
+    "mistral-small-latest",
+    "open-mistral-nemo",
+    "ministral-8b-latest",
+    "ministral-3b-latest",
+    "magistral-medium-latest",
+    "magistral-small-latest",
+    "pixtral-large-latest",
+    "codestral-latest",
+]
+
+
+@app.get("/models/mistral")
+async def list_mistral_models(
+    settings: Annotated[Settings, Depends(get_settings)],
+) -> dict[str, list[dict[str, object]]]:
+    """Catalogue des modeles Mistral. Dynamique si une cle est configuree
+    (GET /v1/models, qui requiert l'auth), sinon repli statique.
+
+    Renvoie {models: [{id}]} (tous accessibles sur le tier gratuit Experiment)."""
+    key = settings.mistral_api_key
+    if not key:
+        return {"models": [{"id": m} for m in _MISTRAL_FALLBACK_MODELS]}
+    try:
+        async with httpx.AsyncClient(timeout=20) as client:
+            response = await client.get(
+                "https://api.mistral.ai/v1/models",
+                headers={"Authorization": f"Bearer {key}"},
+            )
+            response.raise_for_status()
+            data = response.json()
+    except httpx.HTTPError:
+        # Cle invalide / API down : on ne casse pas l'UI, on propose le repli.
+        return {"models": [{"id": m} for m in _MISTRAL_FALLBACK_MODELS]}
+
+    ids = sorted({str(m.get("id")) for m in data.get("data", []) or [] if m.get("id")})
+    if not ids:
+        ids = _MISTRAL_FALLBACK_MODELS
+    return {"models": [{"id": i} for i in ids]}
+
+
+# Repli statique Gemini (juin 2026). Pour l'extraction, prefere un Flash a grand
+# contexte ; `gemini-2.0-flash` a le quota gratuit le plus genereux.
+_GEMINI_FALLBACK_MODELS = [
+    "gemini-2.0-flash",
+    "gemini-2.0-flash-lite",
+    "gemini-2.5-flash",
+    "gemini-2.5-flash-lite",
+    "gemini-2.5-pro",
+    "gemini-1.5-flash",
+    "gemini-1.5-pro",
+]
+
+
+@app.get("/models/gemini")
+async def list_gemini_models(
+    settings: Annotated[Settings, Depends(get_settings)],
+) -> dict[str, list[dict[str, object]]]:
+    """Catalogue des modeles Gemini. Dynamique si une cle est configuree (endpoint
+    OpenAI-compatible /openai/models), sinon repli statique. Renvoie {models:[{id}]}."""
+    key = settings.gemini_api_key
+    if not key:
+        return {"models": [{"id": m} for m in _GEMINI_FALLBACK_MODELS]}
+    try:
+        async with httpx.AsyncClient(timeout=20) as client:
+            response = await client.get(
+                "https://generativelanguage.googleapis.com/v1beta/openai/models",
+                headers={"Authorization": f"Bearer {key}"},
+            )
+            response.raise_for_status()
+            data = response.json()
+    except httpx.HTTPError:
+        return {"models": [{"id": m} for m in _GEMINI_FALLBACK_MODELS]}
+
+    # Les ids peuvent arriver prefixes "models/" → on nettoie pour que la valeur
+    # selectionnee soit directement utilisable dans l'appel chat. On garde les
+    # modeles "gemini-*" (hors embeddings/aqa) pour ne pas noyer la liste.
+    ids: set[str] = set()
+    for m in data.get("data", []) or []:
+        mid = str(m.get("id") or "")
+        if mid.startswith("models/"):
+            mid = mid[len("models/"):]
+        if mid.startswith("gemini-"):
+            ids.add(mid)
+    clean = sorted(ids) if ids else _GEMINI_FALLBACK_MODELS
+    return {"models": [{"id": i} for i in clean]}
+
+
 @app.get("/models/onemin")
 def list_onemin_models() -> dict[str, list[dict[str, object]]]:
     """Catalogue statique des modeles 1min.ai, groupes par fournisseur.

core/pom.xml

@@ -14,7 +14,7 @@
 
     <groupId>com.loremind</groupId>
     <artifactId>loremind-core</artifactId>
-    <version>0.10.0-beta</version>
+    <version>0.11.0-beta</version>
     <name>LoreMind Core</name>
     <description>Backend Core - Architecture Hexagonale</description>
 

core/src/main/java/com/loremind/application/campaigncontext/CampaignAdaptService.java

@@ -1,46 +1,32 @@
 package com.loremind.application.campaigncontext;
 
-import com.loremind.application.generationcontext.CampaignStructuralContextBuilder;
-import com.loremind.application.generationcontext.LoreStructuralContextBuilder;
 import com.loremind.domain.campaigncontext.Campaign;
 import com.loremind.domain.campaigncontext.ports.CampaignPdfAdvisor;
 import com.loremind.domain.campaigncontext.ports.CampaignRepository;
-import com.loremind.domain.generationcontext.CampaignStructuralContext;
-import com.loremind.domain.generationcontext.CampaignStructuralContext.ArcSummary;
-import com.loremind.domain.generationcontext.CampaignStructuralContext.ChapterSummary;
-import com.loremind.domain.generationcontext.CampaignStructuralContext.NpcSummary;
-import com.loremind.domain.generationcontext.CampaignStructuralContext.SceneSummary;
-import com.loremind.domain.generationcontext.LoreStructuralContext;
-import com.loremind.domain.generationcontext.LoreStructuralContext.PageSummary;
 import org.springframework.stereotype.Service;
 
-import java.util.List;
-import java.util.Map;
 import java.util.function.Consumer;
 
 /**
  * Service applicatif : conseils d'adaptation d'un PDF à une campagne existante.
  *
- * <p>Assemble un « brief » de la campagne (structure + PNJ + univers/lore) — la même
+ * <p>Assemble un « brief » de la campagne (structure + PNJ + univers/lore) via
- * matière que le chat de campagne — et délègue la génération streamée au Brain via
+ * {@link CampaignBriefBuilder} et délègue la génération streamée au Brain via
  * {@link CampaignPdfAdvisor}. Ne persiste rien : la sortie est du conseil libre.</p>
  */
 @Service
 public class CampaignAdaptService {
 
     private final CampaignRepository campaignRepository;
-    private final CampaignStructuralContextBuilder campaignContextBuilder;
+    private final CampaignBriefBuilder briefBuilder;
-    private final LoreStructuralContextBuilder loreContextBuilder;
     private final CampaignPdfAdvisor advisor;
 
     public CampaignAdaptService(
             CampaignRepository campaignRepository,
-            CampaignStructuralContextBuilder campaignContextBuilder,
+            CampaignBriefBuilder briefBuilder,
-            LoreStructuralContextBuilder loreContextBuilder,
             CampaignPdfAdvisor advisor) {
         this.campaignRepository = campaignRepository;
-        this.campaignContextBuilder = campaignContextBuilder;
+        this.briefBuilder = briefBuilder;
-        this.loreContextBuilder = loreContextBuilder;
         this.advisor = advisor;
     }
 
@@ -55,64 +41,6 @@ public class CampaignAdaptService {
         Campaign campaign = campaignRepository.findById(campaignId)
                 .orElseThrow(() -> new IllegalArgumentException("Campagne introuvable : " + campaignId));
         advisor.adviseStreaming(
-                pdfBytes, filename, buildBrief(campaign), messagesJson, onToken, onComplete, onError);
+                pdfBytes, filename, briefBuilder.build(campaign), messagesJson, onToken, onComplete, onError);
-    }
-
-    /** Construit un résumé markdown de la campagne (structure + PNJ + lore). */
-    private String buildBrief(Campaign campaign) {
-        CampaignStructuralContext cc = campaignContextBuilder.build(campaign.getId());
-        StringBuilder sb = new StringBuilder();
-
-        sb.append("# Campagne : ").append(cc.campaignName()).append("\n");
-        if (notBlank(cc.campaignDescription())) sb.append(cc.campaignDescription()).append("\n");
-
-        sb.append("\n## Structure (arcs → chapitres → scènes)\n");
-        if (cc.arcs().isEmpty()) {
-            sb.append("_(aucun arc pour le moment)_\n");
-        }
-        for (ArcSummary arc : cc.arcs()) {
-            sb.append("### Arc : ").append(arc.name());
-            if (notBlank(arc.description())) sb.append(" — ").append(arc.description());
-            sb.append("\n");
-            for (ChapterSummary ch : arc.chapters()) {
-                sb.append("- Chapitre : ").append(ch.name());
-                if (notBlank(ch.description())) sb.append(" — ").append(ch.description());
-                sb.append("\n");
-                for (SceneSummary sc : ch.scenes()) {
-                    sb.append("  - Scène : ").append(sc.name());
-                    if (notBlank(sc.description())) sb.append(" — ").append(sc.description());
-                    sb.append("\n");
-                }
-            }
-        }
-
-        if (!cc.npcs().isEmpty()) {
-            sb.append("\n## PNJ existants\n");
-            for (NpcSummary n : cc.npcs()) {
-                sb.append("- ").append(n.name());
-                if (notBlank(n.snippet())) sb.append(" : ").append(n.snippet());
-                sb.append("\n");
-            }
-        }
-
-        if (campaign.isLinkedToLore()) {
-            loreContextBuilder.buildOptional(campaign.getLoreId()).ifPresent(lore -> appendLore(sb, lore));
-        }
-        return sb.toString();
-    }
-
-    private void appendLore(StringBuilder sb, LoreStructuralContext lore) {
-        sb.append("\n## Univers (Lore) : ").append(lore.loreName()).append("\n");
-        if (notBlank(lore.loreDescription())) sb.append(lore.loreDescription()).append("\n");
-        for (Map.Entry<String, List<PageSummary>> entry : lore.folders().entrySet()) {
-            sb.append("### ").append(entry.getKey()).append("\n");
-            for (PageSummary page : entry.getValue()) {
-                sb.append("- ").append(page.title()).append("\n");
-            }
-        }
-    }
-
-    private static boolean notBlank(String s) {
-        return s != null && !s.isBlank();
     }
 }

core/src/main/java/com/loremind/application/campaigncontext/CampaignBriefBuilder.java

@@ -0,0 +1,93 @@
+package com.loremind.application.campaigncontext;
+
+import com.loremind.application.generationcontext.CampaignStructuralContextBuilder;
+import com.loremind.application.generationcontext.LoreStructuralContextBuilder;
+import com.loremind.domain.campaigncontext.Campaign;
+import com.loremind.domain.generationcontext.CampaignStructuralContext;
+import com.loremind.domain.generationcontext.CampaignStructuralContext.ArcSummary;
+import com.loremind.domain.generationcontext.CampaignStructuralContext.ChapterSummary;
+import com.loremind.domain.generationcontext.CampaignStructuralContext.NpcSummary;
+import com.loremind.domain.generationcontext.CampaignStructuralContext.SceneSummary;
+import com.loremind.domain.generationcontext.LoreStructuralContext;
+import com.loremind.domain.generationcontext.LoreStructuralContext.PageSummary;
+import org.springframework.stereotype.Service;
+
+import java.util.List;
+import java.util.Map;
+
+/**
+ * Construit un résumé markdown d'une campagne (structure arcs→chapitres→scènes +
+ * PNJ + univers/lore). Partagé par les fonctions IA qui doivent « voir » la
+ * campagne : conseils d'adaptation PDF (CampaignAdaptService) et ateliers RAG
+ * (NotebookService). Centralisé ici pour une seule source de vérité.
+ */
+@Service
+public class CampaignBriefBuilder {
+
+    private final CampaignStructuralContextBuilder campaignContextBuilder;
+    private final LoreStructuralContextBuilder loreContextBuilder;
+
+    public CampaignBriefBuilder(
+            CampaignStructuralContextBuilder campaignContextBuilder,
+            LoreStructuralContextBuilder loreContextBuilder) {
+        this.campaignContextBuilder = campaignContextBuilder;
+        this.loreContextBuilder = loreContextBuilder;
+    }
+
+    public String build(Campaign campaign) {
+        CampaignStructuralContext cc = campaignContextBuilder.build(campaign.getId());
+        StringBuilder sb = new StringBuilder();
+
+        sb.append("# Campagne : ").append(cc.campaignName()).append("\n");
+        if (notBlank(cc.campaignDescription())) sb.append(cc.campaignDescription()).append("\n");
+
+        sb.append("\n## Structure (arcs → chapitres → scènes)\n");
+        if (cc.arcs().isEmpty()) {
+            sb.append("_(aucun arc pour le moment)_\n");
+        }
+        for (ArcSummary arc : cc.arcs()) {
+            sb.append("### Arc : ").append(arc.name());
+            if (notBlank(arc.description())) sb.append(" — ").append(arc.description());
+            sb.append("\n");
+            for (ChapterSummary ch : arc.chapters()) {
+                sb.append("- Chapitre : ").append(ch.name());
+                if (notBlank(ch.description())) sb.append(" — ").append(ch.description());
+                sb.append("\n");
+                for (SceneSummary sc : ch.scenes()) {
+                    sb.append("  - Scène : ").append(sc.name());
+                    if (notBlank(sc.description())) sb.append(" — ").append(sc.description());
+                    sb.append("\n");
+                }
+            }
+        }
+
+        if (!cc.npcs().isEmpty()) {
+            sb.append("\n## PNJ existants\n");
+            for (NpcSummary n : cc.npcs()) {
+                sb.append("- ").append(n.name());
+                if (notBlank(n.snippet())) sb.append(" : ").append(n.snippet());
+                sb.append("\n");
+            }
+        }
+
+        if (campaign.isLinkedToLore()) {
+            loreContextBuilder.buildOptional(campaign.getLoreId()).ifPresent(lore -> appendLore(sb, lore));
+        }
+        return sb.toString();
+    }
+
+    private void appendLore(StringBuilder sb, LoreStructuralContext lore) {
+        sb.append("\n## Univers (Lore) : ").append(lore.loreName()).append("\n");
+        if (notBlank(lore.loreDescription())) sb.append(lore.loreDescription()).append("\n");
+        for (Map.Entry<String, List<PageSummary>> entry : lore.folders().entrySet()) {
+            sb.append("### ").append(entry.getKey()).append("\n");
+            for (PageSummary page : entry.getValue()) {
+                sb.append("- ").append(page.title()).append("\n");
+            }
+        }
+    }
+
+    private static boolean notBlank(String s) {
+        return s != null && !s.isBlank();
+    }
+}

core/src/main/java/com/loremind/application/campaigncontext/NotebookService.java

@@ -0,0 +1,132 @@
+package com.loremind.application.campaigncontext;
+
+import com.loremind.domain.campaigncontext.Campaign;
+import com.loremind.domain.campaigncontext.Notebook;
+import com.loremind.domain.campaigncontext.NotebookMessage;
+import com.loremind.domain.campaigncontext.NotebookSource;
+import com.loremind.domain.campaigncontext.ports.CampaignRepository;
+import com.loremind.domain.campaigncontext.ports.NotebookIndexer;
+import com.loremind.domain.campaigncontext.ports.NotebookRepository;
+import org.springframework.stereotype.Service;
+
+import java.util.List;
+
+/**
+ * Service d'application des notebooks (atelier RAG) : CRUD, indexation des sources
+ * (déléguée au Brain), conversation persistée, et assemblage du contexte campagne.
+ */
+@Service
+public class NotebookService {
+
+    private final NotebookRepository repository;
+    private final NotebookIndexer indexer;
+    private final CampaignRepository campaignRepository;
+    private final CampaignBriefBuilder briefBuilder;
+
+    public NotebookService(
+            NotebookRepository repository,
+            NotebookIndexer indexer,
+            CampaignRepository campaignRepository,
+            CampaignBriefBuilder briefBuilder) {
+        this.repository = repository;
+        this.indexer = indexer;
+        this.campaignRepository = campaignRepository;
+        this.briefBuilder = briefBuilder;
+    }
+
+    // --- Notebooks ---
+
+    public Notebook createNotebook(String campaignId, String name) {
+        String safeName = (name == null || name.isBlank()) ? "Nouvel atelier" : name.trim();
+        return repository.save(Notebook.builder().campaignId(campaignId).name(safeName).build());
+    }
+
+    public java.util.Optional<Notebook> getNotebook(String id) {
+        return repository.findById(id);
+    }
+
+    public List<Notebook> getNotebooksByCampaign(String campaignId) {
+        return repository.findByCampaignId(campaignId);
+    }
+
+    public Notebook renameNotebook(String id, String name) {
+        Notebook nb = repository.findById(id)
+                .orElseThrow(() -> new IllegalArgumentException("Notebook introuvable: " + id));
+        nb.setName((name == null || name.isBlank()) ? nb.getName() : name.trim());
+        return repository.save(nb);
+    }
+
+    public void deleteNotebook(String id) {
+        // Supprime les vecteurs de chaque source côté Brain (best-effort) avant la BDD.
+        for (NotebookSource s : repository.findSourcesByNotebookId(id)) {
+            indexer.delete(s.getId());
+        }
+        repository.deleteById(id);
+    }
+
+    // --- Sources ---
+
+    public List<NotebookSource> getSources(String notebookId) {
+        return repository.findSourcesByNotebookId(notebookId);
+    }
+
+    /**
+     * Ajoute une source : crée la ligne (INDEXING), lance l'indexation Brain, puis
+     * met à jour (READY + compteurs) ou (FAILED) en cas d'échec — et relaie l'erreur.
+     */
+    public NotebookSource addSource(String notebookId, String filename, byte[] pdfBytes) {
+        if (!repository.existsById(notebookId)) {
+            throw new IllegalArgumentException("Notebook introuvable: " + notebookId);
+        }
+        NotebookSource source = repository.saveSource(NotebookSource.builder()
+                .notebookId(notebookId)
+                .filename(filename != null && !filename.isBlank() ? filename : "source.pdf")
+                .status("INDEXING")
+                .build());
+        try {
+            NotebookIndexer.IndexResult result = indexer.index(source.getId(), pdfBytes, filename);
+            source.setStatus("READY");
+            source.setChunkCount(result.chunks());
+            source.setPageCount(result.pageCount());
+            return repository.saveSource(source);
+        } catch (RuntimeException e) {
+            source.setStatus("FAILED");
+            repository.saveSource(source);
+            throw e;
+        }
+    }
+
+    public void deleteSource(String sourceId) {
+        repository.findSourceById(sourceId).ifPresent(s -> indexer.delete(s.getId()));
+        repository.deleteSourceById(sourceId);
+    }
+
+    public List<String> readySourceIds(String notebookId) {
+        return repository.findSourcesByNotebookId(notebookId).stream()
+                .filter(s -> "READY".equals(s.getStatus()))
+                .map(NotebookSource::getId)
+                .toList();
+    }
+
+    // --- Conversation ---
+
+    public List<NotebookMessage> getMessages(String notebookId) {
+        return repository.findMessagesByNotebookId(notebookId);
+    }
+
+    public NotebookMessage addMessage(String notebookId, String role, String content) {
+        return repository.saveMessage(NotebookMessage.builder()
+                .notebookId(notebookId).role(role).content(content).build());
+    }
+
+    // --- Contexte campagne (oriente l'IA) ---
+
+    /** Brief COMPLET de la campagne (structure arcs/chapitres/scènes + PNJ + lore) :
+     *  l'IA « voit » la campagne, pas seulement son nom. */
+    public String buildContext(String campaignId) {
+        if (campaignId == null) return "";
+        Campaign campaign = campaignRepository.findById(campaignId).orElse(null);
+        if (campaign == null) return "";
+        return briefBuilder.build(campaign);
+    }
+}

core/src/main/java/com/loremind/application/campaigncontext/RandomTableService.java

@@ -0,0 +1,133 @@
+package com.loremind.application.campaigncontext;
+
+import com.loremind.domain.campaigncontext.Campaign;
+import com.loremind.domain.campaigncontext.RandomTable;
+import com.loremind.domain.campaigncontext.RandomTableEntry;
+import com.loremind.domain.campaigncontext.ports.CampaignRepository;
+import com.loremind.domain.campaigncontext.ports.RandomTableGenerator;
+import com.loremind.domain.campaigncontext.ports.RandomTableRepository;
+import com.loremind.domain.gamesystemcontext.ports.GameSystemRepository;
+import org.springframework.stereotype.Service;
+
+import java.util.ArrayList;
+import java.util.List;
+import java.util.Optional;
+
+/**
+ * Service d'application pour les tables aléatoires (campagne).
+ */
+@Service
+public class RandomTableService {
+
+    private final RandomTableRepository repository;
+    private final RandomTableGenerator generator;
+    private final CampaignRepository campaignRepository;
+    private final GameSystemRepository gameSystemRepository;
+
+    public RandomTableService(
+            RandomTableRepository repository,
+            RandomTableGenerator generator,
+            CampaignRepository campaignRepository,
+            GameSystemRepository gameSystemRepository) {
+        this.repository = repository;
+        this.generator = generator;
+        this.campaignRepository = campaignRepository;
+        this.gameSystemRepository = gameSystemRepository;
+    }
+
+    public record TableData(
+            String name,
+            String description,
+            String diceFormula,
+            String icon,
+            List<RandomTableEntry> entries,
+            String campaignId,
+            Integer order
+    ) {}
+
+    public RandomTable createTable(TableData data) {
+        int order = data.order() != null ? data.order() : nextOrderFor(data.campaignId());
+        RandomTable table = RandomTable.builder()
+                .name(data.name())
+                .description(data.description())
+                .diceFormula(data.diceFormula())
+                .icon(data.icon())
+                .entries(copyEntries(data.entries()))
+                .campaignId(data.campaignId())
+                .order(order)
+                .build();
+        return repository.save(table);
+    }
+
+    public Optional<RandomTable> getTableById(String id) {
+        return repository.findById(id);
+    }
+
+    public List<RandomTable> getTablesByCampaignId(String campaignId) {
+        return repository.findByCampaignId(campaignId);
+    }
+
+    public RandomTable updateTable(String id, TableData data) {
+        RandomTable existing = repository.findById(id)
+                .orElseThrow(() -> new IllegalArgumentException("Table aléatoire introuvable: " + id));
+        existing.setName(data.name());
+        existing.setDescription(data.description());
+        existing.setDiceFormula(data.diceFormula());
+        existing.setIcon(data.icon());
+        existing.setEntries(copyEntries(data.entries()));
+        if (data.order() != null) {
+            existing.setOrder(data.order());
+        }
+        return repository.save(existing);
+    }
+
+    public void deleteTable(String id) {
+        repository.deleteById(id);
+    }
+
+    /** Génère une PROPOSITION de table (non persistée) via l'IA, contextualisée campagne. */
+    public RandomTable generateProposal(String campaignId, String description, String diceFormula) {
+        String formula = (diceFormula == null || diceFormula.isBlank()) ? "1d20" : diceFormula;
+        RandomTableGenerator.GeneratedTable g = generator.generate(description, formula, buildContext(campaignId));
+        return RandomTable.builder()
+                .name(g.name())
+                .description(g.description())
+                .diceFormula(formula)
+                .campaignId(campaignId)
+                .entries(g.entries() != null ? new ArrayList<>(g.entries()) : new ArrayList<>())
+                .build();
+    }
+
+    /** Brode un court récit IA sur un résultat tiré (pour la partie). */
+    public String improviseRoll(String campaignId, String tableName, String resultLabel, String resultDetail) {
+        return generator.improvise(tableName, resultLabel, resultDetail, buildContext(campaignId));
+    }
+
+    /** Contexte libre (nom de campagne + description + système) pour orienter l'IA. */
+    private String buildContext(String campaignId) {
+        if (campaignId == null) return "";
+        Campaign campaign = campaignRepository.findById(campaignId).orElse(null);
+        if (campaign == null) return "";
+        StringBuilder sb = new StringBuilder();
+        sb.append("Campagne : ").append(campaign.getName());
+        if (campaign.getDescription() != null && !campaign.getDescription().isBlank()) {
+            sb.append(" — ").append(campaign.getDescription().trim());
+        }
+        if (campaign.getGameSystemId() != null && !campaign.getGameSystemId().isBlank()) {
+            gameSystemRepository.findById(campaign.getGameSystemId())
+                    .ifPresent(gs -> sb.append("\nSystème de jeu : ").append(gs.getName()));
+        }
+        return sb.toString();
+    }
+
+    private static List<RandomTableEntry> copyEntries(List<RandomTableEntry> entries) {
+        return entries != null ? new ArrayList<>(entries) : new ArrayList<>();
+    }
+
+    private int nextOrderFor(String campaignId) {
+        return repository.findByCampaignId(campaignId).stream()
+                .mapToInt(RandomTable::getOrder)
+                .max()
+                .orElse(-1) + 1;
+    }
+}

core/src/main/java/com/loremind/application/generationcontext/SessionStructuralContextBuilder.java

@@ -174,10 +174,9 @@ public class SessionStructuralContextBuilder {
                 .filter(a -> a.getId() != null)
                 .collect(Collectors.toMap(Arc::getId, a -> a));
 
-        boolean anyHub = arcs.stream().anyMatch(a -> a.getType() == ArcType.HUB);
+        // On suit comme "quêtes" les chapitres CONDITIONNELS : ceux d'un arc HUB, ET
-        if (!anyHub) {
+        // ceux d'un arc linéaire qui portent des prérequis. Un chapitre linéaire sans
-            return new HubStatus(List.of(), List.of(), List.of(), activeFlags);
+        // condition reste hors du tableau (sinon tous les chapitres deviendraient des quêtes).
-        }
 
         // Map chapterId -> ProgressionStatus pour ce Playthrough
         Map<String, ProgressionStatus> progressionByChapter = new HashMap<>();
@@ -197,8 +196,10 @@ public class SessionStructuralContextBuilder {
         List<String> lockedTitles = new ArrayList<>();
 
         for (Arc arc : arcs) {
-            if (arc.getType() != ArcType.HUB) continue;
+            boolean isHub = arc.getType() == ArcType.HUB;
             for (Chapter c : chapterRepository.findByArcId(arc.getId())) {
+                boolean hasPrereqs = c.getPrerequisites() != null && !c.getPrerequisites().isEmpty();
+                if (!isHub && !hasPrereqs) continue;  // chapitre linéaire sans condition : ignoré
                 ProgressionStatus prog = progressionByChapter.getOrDefault(c.getId(), ProgressionStatus.NOT_STARTED);
                 QuestStatus status = prerequisiteEvaluator.computeStatus(prog, c.getPrerequisites(), ctx);
                 Arc parent = arcsById.get(c.getArcId());

core/src/main/java/com/loremind/domain/campaigncontext/Notebook.java

@@ -0,0 +1,23 @@
+package com.loremind.domain.campaigncontext;
+
+import lombok.Builder;
+import lombok.Data;
+
+import java.time.LocalDateTime;
+
+/**
+ * Atelier d'adaptation (« notebook ») d'une campagne : une ou plusieurs sources
+ * PDF indexées (RAG) + une conversation, persistés pour y revenir.
+ * <p>
+ * Les SOURCES ({@link NotebookSource}) et les MESSAGES ({@link NotebookMessage})
+ * sont gérés comme entités liées par {@code notebookId} (chargées séparément).
+ */
+@Data
+@Builder
+public class Notebook {
+    private String id;
+    private String name;
+    private String campaignId;
+    private LocalDateTime createdAt;
+    private LocalDateTime updatedAt;
+}

core/src/main/java/com/loremind/domain/campaigncontext/NotebookMessage.java

@@ -0,0 +1,20 @@
+package com.loremind.domain.campaigncontext;
+
+import lombok.Builder;
+import lombok.Data;
+
+import java.time.LocalDateTime;
+
+/**
+ * Un message de la conversation d'un {@link Notebook}. {@code role} = "user" ou
+ * "assistant". Persisté pour recharger l'historique de l'atelier.
+ */
+@Data
+@Builder
+public class NotebookMessage {
+    private String id;
+    private String notebookId;
+    private String role;
+    private String content;
+    private LocalDateTime createdAt;
+}

core/src/main/java/com/loremind/domain/campaigncontext/NotebookSource.java

@@ -0,0 +1,24 @@
+package com.loremind.domain.campaigncontext;
+
+import lombok.Builder;
+import lombok.Data;
+
+import java.time.LocalDateTime;
+
+/**
+ * Une source (PDF) d'un {@link Notebook}. Son {@code id} sert de clé d'indexation
+ * vectorielle côté Brain (les vecteurs vivent sur le volume du Brain).
+ * <p>
+ * {@code status} : INDEXING (en cours), READY (interrogeable), FAILED (échec).
+ */
+@Data
+@Builder
+public class NotebookSource {
+    private String id;
+    private String notebookId;
+    private String filename;
+    private String status;
+    private int chunkCount;
+    private int pageCount;
+    private LocalDateTime createdAt;
+}

core/src/main/java/com/loremind/domain/campaigncontext/RandomTable.java

@@ -0,0 +1,50 @@
+package com.loremind.domain.campaigncontext;
+
+import lombok.Builder;
+import lombok.Data;
+
+import java.time.LocalDateTime;
+import java.util.ArrayList;
+import java.util.List;
+
+/**
+ * Table aléatoire d'une campagne : on jette un dé ({@code diceFormula}) et la
+ * valeur tombée désigne une {@link RandomTableEntry} (par sa plage) → un résultat.
+ * <p>
+ * Outil MJ classique (rencontres, butin, complications, noms…). Le JET lui-même
+ * est effectué côté client (instantané, comme le panneau de dés) ; le domaine ne
+ * fait que stocker la table et ses entrées. Scope campagne (cross-aggregate via ID).
+ */
+@Data
+@Builder
+public class RandomTable {
+
+    private String id;
+    private String name;
+
+    /** Description libre (à quoi sert la table). Nullable. */
+    private String description;
+
+    /** Formule du dé à lancer : "1d20", "2d6", "d100"… */
+    private String diceFormula;
+
+    /** Clé d'icône (lucide) pour la sidebar/fiche. Nullable. */
+    private String icon;
+
+    /** Référence vers la Campaign parente (cross-aggregate via ID). */
+    private String campaignId;
+
+    /** Ordre d'affichage dans la liste des tables de la campagne. */
+    private int order;
+
+    /** Entrées ordonnées (par plage de jet). Jamais null après construction. */
+    private List<RandomTableEntry> entries;
+
+    private LocalDateTime createdAt;
+    private LocalDateTime updatedAt;
+
+    public List<RandomTableEntry> getEntries() {
+        if (entries == null) entries = new ArrayList<>();
+        return entries;
+    }
+}

core/src/main/java/com/loremind/domain/campaigncontext/RandomTableEntry.java

@@ -0,0 +1,29 @@
+package com.loremind.domain.campaigncontext;
+
+import lombok.Builder;
+import lombok.Data;
+
+/**
+ * Une entrée d'une {@link RandomTable} : une PLAGE de jet (minRoll..maxRoll, bornes
+ * incluses) qui mappe vers un résultat. Les plages permettent les tables PONDÉRÉES
+ * (un résultat couvrant 1–10 est plus probable qu'un couvrant 11–12).
+ * <p>
+ * Value object possédé par la table (pas d'identité propre côté domaine) : à chaque
+ * mise à jour, les entrées sont remplacées en bloc.
+ */
+@Data
+@Builder
+public class RandomTableEntry {
+
+    /** Borne basse du jet (incluse). */
+    private int minRoll;
+
+    /** Borne haute du jet (incluse). Pour une entrée unitaire, min == max. */
+    private int maxRoll;
+
+    /** Résultat court affiché (ex. "Embuscade de gobelins"). */
+    private String label;
+
+    /** Détail markdown : « ce que c'est » (effet, description). Nullable. */
+    private String detail;
+}

core/src/main/java/com/loremind/domain/campaigncontext/ports/NotebookChatStreamer.java

@@ -0,0 +1,36 @@
+package com.loremind.domain.campaigncontext.ports;
+
+import java.util.List;
+import java.util.function.Consumer;
+
+/**
+ * Port de sortie : chat ANCRÉ (RAG) sur les sources d'un notebook, streamé.
+ * Le Brain récupère les passages pertinents puis streame la réponse token par token.
+ */
+public interface NotebookChatStreamer {
+
+    /** Un message de la conversation transmis au Brain. */
+    record Msg(String role, String content) {}
+
+    /** Avancement de l'analyse approfondie (lecture du document par lots). */
+    record Progress(int current, int total) {}
+
+    /**
+     * Streame la réponse ancrée sur les sources. Les callbacks sont invoqués au fil
+     * de l'eau : {@code onToken} par fragment, {@code onProgress} (mode approfondi
+     * uniquement) pendant la lecture du document, {@code onDone} à la fin,
+     * {@code onError} en cas d'échec.
+     *
+     * @param deep true = analyse approfondie (map-reduce sur tout le document) ;
+     *             false = chat RAG (top-k).
+     */
+    void stream(
+            List<String> sourceIds,
+            List<Msg> messages,
+            String context,
+            boolean deep,
+            Consumer<String> onToken,
+            Consumer<Progress> onProgress,
+            Runnable onDone,
+            Consumer<Throwable> onError);
+}

core/src/main/java/com/loremind/domain/campaigncontext/ports/NotebookException.java

@@ -0,0 +1,15 @@
+package com.loremind.domain.campaigncontext.ports;
+
+/**
+ * Échec d'indexation/chat d'un notebook (Brain injoignable, erreur du modèle…).
+ * Mappée en HTTP 502 par le contrôleur.
+ */
+public class NotebookException extends RuntimeException {
+    public NotebookException(String message) {
+        super(message);
+    }
+
+    public NotebookException(String message, Throwable cause) {
+        super(message, cause);
+    }
+}

core/src/main/java/com/loremind/domain/campaigncontext/ports/NotebookIndexer.java

@@ -0,0 +1,17 @@
+package com.loremind.domain.campaigncontext.ports;
+
+/**
+ * Port de sortie : indexation RAG d'une source de notebook (déléguée au Brain).
+ * Les vecteurs vivent côté Brain, keyés par {@code sourceId}.
+ */
+public interface NotebookIndexer {
+
+    /** Récapitulatif d'indexation renvoyé par le Brain. */
+    record IndexResult(int chunks, int pageCount, int ocrPageCount) {}
+
+    /** Indexe une source (extraction + embeddings + stockage vectoriel). */
+    IndexResult index(String sourceId, byte[] pdfBytes, String filename);
+
+    /** Supprime les vecteurs d'une source (au DELETE d'une source/notebook). */
+    void delete(String sourceId);
+}

core/src/main/java/com/loremind/domain/campaigncontext/ports/NotebookRepository.java

@@ -0,0 +1,32 @@
+package com.loremind.domain.campaigncontext.ports;
+
+import com.loremind.domain.campaigncontext.Notebook;
+import com.loremind.domain.campaigncontext.NotebookMessage;
+import com.loremind.domain.campaigncontext.NotebookSource;
+
+import java.util.List;
+import java.util.Optional;
+
+/**
+ * Port de sortie pour la persistance des notebooks (atelier), de leurs sources
+ * et de leur conversation. Port unique (3 agrégats liés) pour rester compact.
+ */
+public interface NotebookRepository {
+
+    // --- Notebook ---
+    Notebook save(Notebook notebook);
+    Optional<Notebook> findById(String id);
+    List<Notebook> findByCampaignId(String campaignId);
+    void deleteById(String id);
+    boolean existsById(String id);
+
+    // --- Sources ---
+    NotebookSource saveSource(NotebookSource source);
+    Optional<NotebookSource> findSourceById(String id);
+    List<NotebookSource> findSourcesByNotebookId(String notebookId);
+    void deleteSourceById(String id);
+
+    // --- Messages (conversation) ---
+    NotebookMessage saveMessage(NotebookMessage message);
+    List<NotebookMessage> findMessagesByNotebookId(String notebookId);
+}

core/src/main/java/com/loremind/domain/campaigncontext/ports/RandomTableGenerationException.java

@@ -0,0 +1,15 @@
+package com.loremind.domain.campaigncontext.ports;
+
+/**
+ * Échec de génération/improvisation IA d'une table (Brain injoignable, erreur du
+ * modèle, réponse inexploitable…). Mappée en HTTP 502 par le contrôleur.
+ */
+public class RandomTableGenerationException extends RuntimeException {
+    public RandomTableGenerationException(String message) {
+        super(message);
+    }
+
+    public RandomTableGenerationException(String message, Throwable cause) {
+        super(message, cause);
+    }
+}

core/src/main/java/com/loremind/domain/campaigncontext/ports/RandomTableGenerator.java

@@ -0,0 +1,26 @@
+package com.loremind.domain.campaigncontext.ports;
+
+import com.loremind.domain.campaigncontext.RandomTableEntry;
+
+import java.util.List;
+
+/**
+ * Port de sortie : génération IA d'une table aléatoire et improvisation narrative
+ * sur un résultat tiré. Implémenté par un client du Brain (service IA Python).
+ */
+public interface RandomTableGenerator {
+
+    /** Table proposée (non persistée) à partir d'une description + formule de dé. */
+    record GeneratedTable(String name, String description, List<RandomTableEntry> entries) {}
+
+    /**
+     * Génère une proposition de table couvrant la formule de dé, sur le sujet
+     * donné, en s'appuyant sur le contexte (campagne, système…) s'il est fourni.
+     */
+    GeneratedTable generate(String description, String diceFormula, String context);
+
+    /**
+     * Brode un court récit (2-3 phrases) sur un résultat tiré, pour lancer la scène.
+     */
+    String improvise(String tableName, String resultLabel, String resultDetail, String context);
+}

core/src/main/java/com/loremind/domain/campaigncontext/ports/RandomTableRepository.java

@@ -0,0 +1,22 @@
+package com.loremind.domain.campaigncontext.ports;
+
+import com.loremind.domain.campaigncontext.RandomTable;
+
+import java.util.List;
+import java.util.Optional;
+
+/**
+ * Port de sortie pour la persistance des {@link RandomTable}.
+ */
+public interface RandomTableRepository {
+
+    RandomTable save(RandomTable table);
+
+    Optional<RandomTable> findById(String id);
+
+    List<RandomTable> findByCampaignId(String campaignId);
+
+    void deleteById(String id);
+
+    boolean existsById(String id);
+}

core/src/main/java/com/loremind/infrastructure/ai/BrainCampaignImportClient.java

@@ -90,8 +90,12 @@ public class BrainCampaignImportClient implements CampaignPdfImporter {
             }
         } catch (Exception e) {
             if (!terminated[0]) {
+                // On EXPOSE la cause réelle (type + message) : sans ça, le diagnostic
+                // est impossible (timeout WebClient, connexion coupée, réponse non-2xx…).
+                String cause = e.getClass().getSimpleName()
+                        + (e.getMessage() != null ? " — " + e.getMessage() : "");
                 onError.accept(new CampaignImportException(
-                        "Erreur lors du streaming d'import depuis le Brain.", e));
+                        "Erreur lors du streaming d'import depuis le Brain : " + cause, e));
             }
         }
     }

core/src/main/java/com/loremind/infrastructure/ai/BrainNotebookChatClient.java

@@ -0,0 +1,139 @@
+package com.loremind.infrastructure.ai;
+
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.loremind.domain.campaigncontext.ports.NotebookChatStreamer;
+import com.loremind.domain.campaigncontext.ports.NotebookException;
+import org.springframework.beans.factory.annotation.Value;
+import org.springframework.core.ParameterizedTypeReference;
+import org.springframework.http.MediaType;
+import org.springframework.http.codec.ServerSentEvent;
+import org.springframework.stereotype.Component;
+import org.springframework.web.reactive.function.client.WebClient;
+import reactor.core.publisher.Flux;
+
+import java.time.Duration;
+import java.util.LinkedHashMap;
+import java.util.List;
+import java.util.Map;
+import java.util.function.Consumer;
+import java.util.stream.Collectors;
+
+/**
+ * Adapter de sortie : relaie le chat ANCRÉ (RAG) du Brain via SSE (WebClient).
+ * Pattern identique aux imports streamés (cf. BrainRulesImportClient).
+ */
+@Component
+public class BrainNotebookChatClient implements NotebookChatStreamer {
+
+    private static final String PATH = "/chat/notebook/stream";
+    private static final String DEEP_PATH = "/chat/notebook/deep/stream";
+    private static final ParameterizedTypeReference<ServerSentEvent<String>> SSE_STRING_TYPE =
+            new ParameterizedTypeReference<>() {};
+
+    private final WebClient webClient;
+    private final ObjectMapper objectMapper;
+    private final long timeoutSeconds;
+
+    public BrainNotebookChatClient(
+            WebClient.Builder webClientBuilder,
+            ObjectMapper objectMapper,
+            @Value("${brain.base-url}") String baseUrl,
+            @Value("${brain.import-timeout-seconds:600}") long timeoutSeconds) {
+        this.webClient = webClientBuilder.baseUrl(baseUrl).build();
+        this.objectMapper = objectMapper;
+        this.timeoutSeconds = timeoutSeconds;
+    }
+
+    @Override
+    public void stream(
+            List<String> sourceIds,
+            List<Msg> messages,
+            String context,
+            boolean deep,
+            Consumer<String> onToken,
+            Consumer<Progress> onProgress,
+            Runnable onDone,
+            Consumer<Throwable> onError) {
+
+        Map<String, Object> payload = new LinkedHashMap<>();
+        payload.put("source_ids", sourceIds);
+        payload.put("messages", messages.stream()
+                .map(m -> Map.<String, Object>of("role", m.role(), "content", m.content()))
+                .collect(Collectors.toList()));
+        payload.put("context", context == null ? "" : context);
+
+        Flux<ServerSentEvent<String>> flux = webClient.post()
+                .uri(deep ? DEEP_PATH : PATH)
+                .contentType(MediaType.APPLICATION_JSON)
+                .accept(MediaType.TEXT_EVENT_STREAM)
+                .bodyValue(payload)
+                .retrieve()
+                .bodyToFlux(SSE_STRING_TYPE);
+
+        boolean[] terminated = {false};
+        try {
+            flux
+                .timeout(Duration.ofSeconds(timeoutSeconds))
+                .doOnNext(sse -> handleEvent(sse, terminated, onToken, onProgress, onDone, onError))
+                .blockLast();
+            if (!terminated[0]) {
+                onDone.run();  // flux terminé sans event done explicite
+            }
+        } catch (Exception e) {
+            if (!terminated[0]) {
+                String cause = e.getClass().getSimpleName()
+                        + (e.getMessage() != null ? " — " + e.getMessage() : "");
+                onError.accept(new NotebookException(
+                        "Erreur lors du streaming du chat depuis le Brain : " + cause, e));
+            }
+        }
+    }
+
+    private void handleEvent(
+            ServerSentEvent<String> sse,
+            boolean[] terminated,
+            Consumer<String> onToken,
+            Consumer<Progress> onProgress,
+            Runnable onDone,
+            Consumer<Throwable> onError) {
+
+        String event = sse.event();
+        String data = sse.data() == null ? "" : sse.data();
+        if ("token".equals(event)) {
+            String token = readField(data, "token");
+            if (token != null && !token.isEmpty()) onToken.accept(token);
+        } else if ("progress".equals(event)) {
+            onProgress.accept(new Progress(readInt(data, "current"), readInt(data, "total")));
+        } else if ("done".equals(event)) {
+            terminated[0] = true;
+            onDone.run();
+        } else if ("error".equals(event)) {
+            terminated[0] = true;
+            onError.accept(new NotebookException("Le Brain a signalé une erreur : " + readMessage(data)));
+        }
+    }
+
+    private int readInt(String data, String field) {
+        try {
+            JsonNode node = objectMapper.readTree(data);
+            return node.hasNonNull(field) ? node.get(field).asInt() : 0;
+        } catch (Exception e) {
+            return 0;
+        }
+    }
+
+    private String readField(String data, String field) {
+        try {
+            JsonNode node = objectMapper.readTree(data);
+            return node.hasNonNull(field) ? node.get(field).asText() : null;
+        } catch (Exception e) {
+            return null;
+        }
+    }
+
+    private String readMessage(String data) {
+        String msg = readField(data, "message");
+        return msg != null ? msg : data;
+    }
+}

core/src/main/java/com/loremind/infrastructure/ai/BrainNotebookIndexClient.java

@@ -0,0 +1,100 @@
+package com.loremind.infrastructure.ai;
+
+import com.fasterxml.jackson.annotation.JsonProperty;
+import com.loremind.domain.campaigncontext.ports.NotebookException;
+import com.loremind.domain.campaigncontext.ports.NotebookIndexer;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+import org.springframework.beans.factory.annotation.Qualifier;
+import org.springframework.beans.factory.annotation.Value;
+import org.springframework.core.io.ByteArrayResource;
+import org.springframework.http.HttpEntity;
+import org.springframework.http.HttpHeaders;
+import org.springframework.http.MediaType;
+import org.springframework.stereotype.Component;
+import org.springframework.util.LinkedMultiValueMap;
+import org.springframework.util.MultiValueMap;
+import org.springframework.web.client.ResourceAccessException;
+import org.springframework.web.client.RestClientResponseException;
+import org.springframework.web.client.RestTemplate;
+
+/**
+ * Adapter de sortie : indexe une source de notebook via le Brain (multipart,
+ * one-shot bloquant — l'indexation d'un livre peut prendre du temps, d'où le
+ * RestTemplate à timeout long {@code brainImportRestTemplate}).
+ */
+@Component
+public class BrainNotebookIndexClient implements NotebookIndexer {
+
+    private static final Logger LOG = LoggerFactory.getLogger(BrainNotebookIndexClient.class);
+    private static final String INDEX_PATH = "/index/notebook-source";
+
+    private final RestTemplate restTemplate;
+    private final String baseUrl;
+
+    public BrainNotebookIndexClient(
+            @Qualifier("brainImportRestTemplate") RestTemplate restTemplate,
+            @Value("${brain.base-url}") String baseUrl) {
+        this.restTemplate = restTemplate;
+        this.baseUrl = baseUrl;
+    }
+
+    @Override
+    public IndexResult index(String sourceId, byte[] pdfBytes, String filename) {
+        HttpHeaders headers = new HttpHeaders();
+        headers.setContentType(MediaType.MULTIPART_FORM_DATA);
+
+        MultiValueMap<String, Object> body = new LinkedMultiValueMap<>();
+        body.add("source_id", sourceId);
+        body.add("file", filePart(pdfBytes, filename));
+        HttpEntity<MultiValueMap<String, Object>> entity = new HttpEntity<>(body, headers);
+
+        try {
+            IndexResponse resp = restTemplate.postForObject(
+                    baseUrl + INDEX_PATH, entity, IndexResponse.class);
+            if (resp == null) {
+                throw new NotebookException("Le Brain a renvoyé une réponse vide à l'indexation.");
+            }
+            return new IndexResult(resp.chunks, resp.pageCount, resp.ocrPageCount);
+        } catch (ResourceAccessException e) {
+            throw new NotebookException("Le Brain est injoignable (timeout ou arrêté).", e);
+        } catch (RestClientResponseException e) {
+            throw new NotebookException(
+                    "Le Brain a répondu HTTP " + e.getStatusCode().value()
+                            + " : " + e.getResponseBodyAsString(), e);
+        } catch (NotebookException e) {
+            throw e;
+        } catch (Exception e) {
+            throw new NotebookException("Erreur inattendue lors de l'indexation via le Brain.", e);
+        }
+    }
+
+    @Override
+    public void delete(String sourceId) {
+        // Best-effort : si le Brain est down, on ne bloque pas la suppression côté Core
+        // (les vecteurs orphelins seront simplement ignorés / nettoyables plus tard).
+        try {
+            restTemplate.delete(baseUrl + INDEX_PATH + "/" + sourceId);
+        } catch (Exception e) {
+            LOG.warn("Suppression des vecteurs de la source {} échouée (ignorée) : {}", sourceId, e.getMessage());
+        }
+    }
+
+    private ByteArrayResource filePart(byte[] pdfBytes, String filename) {
+        return new ByteArrayResource(pdfBytes) {
+            @Override
+            public String getFilename() {
+                return (filename == null || filename.isBlank()) ? "source.pdf" : filename;
+            }
+        };
+    }
+
+    /** Réponse JSON du Brain (snake_case). */
+    private static class IndexResponse {
+        public int chunks;
+        @JsonProperty("page_count")
+        public int pageCount;
+        @JsonProperty("ocr_page_count")
+        public int ocrPageCount;
+    }
+}

core/src/main/java/com/loremind/infrastructure/ai/BrainRandomTableClient.java

@@ -0,0 +1,120 @@
+package com.loremind.infrastructure.ai;
+
+import com.loremind.domain.campaigncontext.RandomTableEntry;
+import com.loremind.domain.campaigncontext.ports.RandomTableGenerationException;
+import com.loremind.domain.campaigncontext.ports.RandomTableGenerator;
+import org.springframework.beans.factory.annotation.Value;
+import org.springframework.http.HttpEntity;
+import org.springframework.http.HttpHeaders;
+import org.springframework.http.MediaType;
+import org.springframework.stereotype.Component;
+import org.springframework.web.client.ResourceAccessException;
+import org.springframework.web.client.RestClientResponseException;
+import org.springframework.web.client.RestTemplate;
+
+import java.util.ArrayList;
+import java.util.LinkedHashMap;
+import java.util.List;
+import java.util.Map;
+
+/**
+ * Adapter de sortie : implémente {@link RandomTableGenerator} en appelant le Brain
+ * (POST one-shot, RestTemplate). Le secret inter-service est injecté par l'intercepteur
+ * du bean {@code brainRestTemplate}.
+ */
+@Component
+public class BrainRandomTableClient implements RandomTableGenerator {
+
+    private static final String GENERATE_PATH = "/generate/random-table";
+    private static final String IMPROVISE_PATH = "/improvise/table-roll";
+
+    private final RestTemplate restTemplate;
+    private final String baseUrl;
+
+    public BrainRandomTableClient(
+            RestTemplate restTemplate,
+            @Value("${brain.base-url}") String baseUrl) {
+        this.restTemplate = restTemplate;
+        this.baseUrl = baseUrl;
+    }
+
+    @Override
+    public GeneratedTable generate(String description, String diceFormula, String context) {
+        Map<String, Object> req = new LinkedHashMap<>();
+        req.put("description", description == null ? "" : description);
+        req.put("dice_formula", diceFormula == null ? "1d20" : diceFormula);
+        req.put("context", context == null ? "" : context);
+
+        Map<String, Object> resp = post(GENERATE_PATH, req);
+        if (resp == null) {
+            throw new RandomTableGenerationException("Le Brain a renvoyé une réponse vide.");
+        }
+        List<RandomTableEntry> entries = new ArrayList<>();
+        Object rawEntries = resp.get("entries");
+        if (rawEntries instanceof List<?> list) {
+            for (Object item : list) {
+                if (!(item instanceof Map<?, ?> m)) continue;
+                Integer min = asInt(m.get("min_roll"));
+                Integer max = asInt(m.get("max_roll"));
+                String label = asString(m.get("label"));
+                if (min == null || max == null || label == null || label.isBlank()) continue;
+                entries.add(RandomTableEntry.builder()
+                        .minRoll(min)
+                        .maxRoll(Math.max(min, max))
+                        .label(label)
+                        .detail(asString(m.get("detail")))
+                        .build());
+            }
+        }
+        if (entries.isEmpty()) {
+            throw new RandomTableGenerationException("Aucune entrée générée — réessaie ou reformule.");
+        }
+        String name = asString(resp.get("name"));
+        return new GeneratedTable(
+                name != null && !name.isBlank() ? name : description,
+                asString(resp.get("description")),
+                entries);
+    }
+
+    @Override
+    public String improvise(String tableName, String resultLabel, String resultDetail, String context) {
+        Map<String, Object> req = new LinkedHashMap<>();
+        req.put("table_name", tableName == null ? "" : tableName);
+        req.put("result_label", resultLabel == null ? "" : resultLabel);
+        req.put("result_detail", resultDetail == null ? "" : resultDetail);
+        req.put("context", context == null ? "" : context);
+
+        Map<String, Object> resp = post(IMPROVISE_PATH, req);
+        String narration = resp != null ? asString(resp.get("narration")) : null;
+        return narration != null ? narration : "";
+    }
+
+    @SuppressWarnings("unchecked")
+    private Map<String, Object> post(String path, Map<String, Object> body) {
+        HttpHeaders headers = new HttpHeaders();
+        headers.setContentType(MediaType.APPLICATION_JSON);
+        HttpEntity<Map<String, Object>> entity = new HttpEntity<>(body, headers);
+        try {
+            return restTemplate.postForObject(baseUrl + path, entity, Map.class);
+        } catch (ResourceAccessException e) {
+            throw new RandomTableGenerationException("Le Brain est injoignable (timeout ou arrêté).", e);
+        } catch (RestClientResponseException e) {
+            throw new RandomTableGenerationException(
+                    "Le Brain a répondu HTTP " + e.getStatusCode().value() + " : " + e.getResponseBodyAsString(), e);
+        } catch (Exception e) {
+            throw new RandomTableGenerationException("Erreur inattendue lors de l'appel au Brain.", e);
+        }
+    }
+
+    private static Integer asInt(Object o) {
+        if (o instanceof Number n) return n.intValue();
+        if (o instanceof String s) {
+            try { return Integer.parseInt(s.trim()); } catch (NumberFormatException ignored) { return null; }
+        }
+        return null;
+    }
+
+    private static String asString(Object o) {
+        return o != null ? o.toString() : null;
+    }
+}

core/src/main/java/com/loremind/infrastructure/ai/BrainRulesImportClient.java

@@ -148,8 +148,13 @@ public class BrainRulesImportClient implements RulesPdfImporter {
             }
         } catch (Exception e) {
             if (!terminated[0]) {
+                // On EXPOSE la cause réelle (type + message) : sans ça, l'UI n'a qu'un
+                // message générique et le diagnostic est impossible (timeout WebClient,
+                // connexion coupée, réponse non-2xx du Brain, etc.).
+                String cause = e.getClass().getSimpleName()
+                        + (e.getMessage() != null ? " — " + e.getMessage() : "");
                 onError.accept(new RulesImportException(
-                        "Erreur lors du streaming d'import depuis le Brain.", e));
+                        "Erreur lors du streaming d'import depuis le Brain : " + cause, e));
             }
         }
     }

core/src/main/java/com/loremind/infrastructure/persistence/entity/NotebookJpaEntity.java

@@ -0,0 +1,47 @@
+package com.loremind.infrastructure.persistence.entity;
+
+import jakarta.persistence.*;
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+import java.time.LocalDateTime;
+
+@Entity
+@Table(name = "notebooks", indexes = {
+        @Index(name = "idx_notebooks_campaign_id", columnList = "campaign_id")
+})
+@Data
+@Builder
+@NoArgsConstructor
+@AllArgsConstructor
+public class NotebookJpaEntity {
+
+    @Id
+    @GeneratedValue(strategy = GenerationType.IDENTITY)
+    private Long id;
+
+    @Column(nullable = false)
+    private String name;
+
+    @Column(name = "campaign_id", nullable = false)
+    private Long campaignId;
+
+    @Column(name = "created_at", nullable = false, updatable = false)
+    private LocalDateTime createdAt;
+
+    @Column(name = "updated_at", nullable = false)
+    private LocalDateTime updatedAt;
+
+    @PrePersist
+    protected void onCreate() {
+        createdAt = LocalDateTime.now();
+        updatedAt = LocalDateTime.now();
+    }
+
+    @PreUpdate
+    protected void onUpdate() {
+        updatedAt = LocalDateTime.now();
+    }
+}

core/src/main/java/com/loremind/infrastructure/persistence/entity/NotebookMessageJpaEntity.java

@@ -0,0 +1,41 @@
+package com.loremind.infrastructure.persistence.entity;
+
+import jakarta.persistence.*;
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+import java.time.LocalDateTime;
+
+@Entity
+@Table(name = "notebook_messages", indexes = {
+        @Index(name = "idx_notebook_messages_notebook_id", columnList = "notebook_id")
+})
+@Data
+@Builder
+@NoArgsConstructor
+@AllArgsConstructor
+public class NotebookMessageJpaEntity {
+
+    @Id
+    @GeneratedValue(strategy = GenerationType.IDENTITY)
+    private Long id;
+
+    @Column(name = "notebook_id", nullable = false)
+    private Long notebookId;
+
+    @Column(nullable = false, length = 16)
+    private String role;
+
+    @Column(columnDefinition = "TEXT", nullable = false)
+    private String content;
+
+    @Column(name = "created_at", nullable = false, updatable = false)
+    private LocalDateTime createdAt;
+
+    @PrePersist
+    protected void onCreate() {
+        if (createdAt == null) createdAt = LocalDateTime.now();
+    }
+}

core/src/main/java/com/loremind/infrastructure/persistence/entity/NotebookSourceJpaEntity.java

@@ -0,0 +1,47 @@
+package com.loremind.infrastructure.persistence.entity;
+
+import jakarta.persistence.*;
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+import java.time.LocalDateTime;
+
+@Entity
+@Table(name = "notebook_sources", indexes = {
+        @Index(name = "idx_notebook_sources_notebook_id", columnList = "notebook_id")
+})
+@Data
+@Builder
+@NoArgsConstructor
+@AllArgsConstructor
+public class NotebookSourceJpaEntity {
+
+    @Id
+    @GeneratedValue(strategy = GenerationType.IDENTITY)
+    private Long id;
+
+    @Column(name = "notebook_id", nullable = false)
+    private Long notebookId;
+
+    @Column(nullable = false)
+    private String filename;
+
+    @Column(nullable = false, length = 16)
+    private String status;
+
+    @Column(name = "chunk_count", nullable = false)
+    private int chunkCount;
+
+    @Column(name = "page_count", nullable = false)
+    private int pageCount;
+
+    @Column(name = "created_at", nullable = false, updatable = false)
+    private LocalDateTime createdAt;
+
+    @PrePersist
+    protected void onCreate() {
+        if (createdAt == null) createdAt = LocalDateTime.now();
+    }
+}

core/src/main/java/com/loremind/infrastructure/persistence/entity/RandomTableEntryJpaEntity.java

@@ -0,0 +1,51 @@
+package com.loremind.infrastructure.persistence.entity;
+
+import jakarta.persistence.*;
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.EqualsAndHashCode;
+import lombok.NoArgsConstructor;
+import lombok.ToString;
+
+/**
+ * Entité JPA d'une entrée de table aléatoire (enfant de {@link RandomTableJpaEntity}).
+ * Ordonnée par {@code position}. La référence parente est exclue de toString/equals
+ * pour éviter les récursions infinies (relation bidirectionnelle).
+ */
+@Entity
+@Table(name = "random_table_entries", indexes = {
+        @Index(name = "idx_random_table_entries_table_id", columnList = "random_table_id")
+})
+@Data
+@Builder
+@NoArgsConstructor
+@AllArgsConstructor
+public class RandomTableEntryJpaEntity {
+
+    @Id
+    @GeneratedValue(strategy = GenerationType.IDENTITY)
+    private Long id;
+
+    @Column(name = "min_roll", nullable = false)
+    private int minRoll;
+
+    @Column(name = "max_roll", nullable = false)
+    private int maxRoll;
+
+    @Column(nullable = false)
+    private String label;
+
+    @Column(columnDefinition = "TEXT")
+    private String detail;
+
+    /** Position d'affichage dans la table (ordre des entrées). */
+    @Column(nullable = false)
+    private int position;
+
+    @ManyToOne(optional = false, fetch = FetchType.LAZY)
+    @JoinColumn(name = "random_table_id", nullable = false)
+    @ToString.Exclude
+    @EqualsAndHashCode.Exclude
+    private RandomTableJpaEntity randomTable;
+}

core/src/main/java/com/loremind/infrastructure/persistence/entity/RandomTableJpaEntity.java

@@ -0,0 +1,73 @@
+package com.loremind.infrastructure.persistence.entity;
+
+import jakarta.persistence.*;
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+import java.time.LocalDateTime;
+import java.util.ArrayList;
+import java.util.List;
+
+/**
+ * Entité JPA d'une table aléatoire (parent) avec ses entrées ordonnées (enfants).
+ * Tables créées automatiquement par Hibernate (ddl-auto=update).
+ */
+@Entity
+@Table(name = "random_tables")
+@Data
+@Builder
+@NoArgsConstructor
+@AllArgsConstructor
+public class RandomTableJpaEntity {
+
+    @Id
+    @GeneratedValue(strategy = GenerationType.IDENTITY)
+    private Long id;
+
+    @Column(nullable = false)
+    private String name;
+
+    @Column(columnDefinition = "TEXT")
+    private String description;
+
+    @Column(name = "dice_formula", nullable = false, length = 32)
+    private String diceFormula;
+
+    @Column(length = 64)
+    private String icon;
+
+    @Column(name = "campaign_id", nullable = false)
+    private Long campaignId;
+
+    @Column(name = "\"order\"", nullable = false)
+    private int order;
+
+    @OneToMany(
+            mappedBy = "randomTable",
+            cascade = CascadeType.ALL,
+            orphanRemoval = true,
+            fetch = FetchType.LAZY
+    )
+    @OrderBy("position ASC")
+    @Builder.Default
+    private List<RandomTableEntryJpaEntity> entries = new ArrayList<>();
+
+    @Column(name = "created_at", nullable = false, updatable = false)
+    private LocalDateTime createdAt;
+
+    @Column(name = "updated_at", nullable = false)
+    private LocalDateTime updatedAt;
+
+    @PrePersist
+    protected void onCreate() {
+        createdAt = LocalDateTime.now();
+        updatedAt = LocalDateTime.now();
+    }
+
+    @PreUpdate
+    protected void onUpdate() {
+        updatedAt = LocalDateTime.now();
+    }
+}

core/src/main/java/com/loremind/infrastructure/persistence/jpa/NotebookJpaRepository.java

@@ -0,0 +1,12 @@
+package com.loremind.infrastructure.persistence.jpa;
+
+import com.loremind.infrastructure.persistence.entity.NotebookJpaEntity;
+import org.springframework.data.jpa.repository.JpaRepository;
+import org.springframework.stereotype.Repository;
+
+import java.util.List;
+
+@Repository
+public interface NotebookJpaRepository extends JpaRepository<NotebookJpaEntity, Long> {
+    List<NotebookJpaEntity> findByCampaignIdOrderByUpdatedAtDesc(Long campaignId);
+}

core/src/main/java/com/loremind/infrastructure/persistence/jpa/NotebookMessageJpaRepository.java

@@ -0,0 +1,13 @@
+package com.loremind.infrastructure.persistence.jpa;
+
+import com.loremind.infrastructure.persistence.entity.NotebookMessageJpaEntity;
+import org.springframework.data.jpa.repository.JpaRepository;
+import org.springframework.stereotype.Repository;
+
+import java.util.List;
+
+@Repository
+public interface NotebookMessageJpaRepository extends JpaRepository<NotebookMessageJpaEntity, Long> {
+    List<NotebookMessageJpaEntity> findByNotebookIdOrderByCreatedAtAsc(Long notebookId);
+    void deleteByNotebookId(Long notebookId);
+}

core/src/main/java/com/loremind/infrastructure/persistence/jpa/NotebookSourceJpaRepository.java

@@ -0,0 +1,13 @@
+package com.loremind.infrastructure.persistence.jpa;
+
+import com.loremind.infrastructure.persistence.entity.NotebookSourceJpaEntity;
+import org.springframework.data.jpa.repository.JpaRepository;
+import org.springframework.stereotype.Repository;
+
+import java.util.List;
+
+@Repository
+public interface NotebookSourceJpaRepository extends JpaRepository<NotebookSourceJpaEntity, Long> {
+    List<NotebookSourceJpaEntity> findByNotebookIdOrderByCreatedAtAsc(Long notebookId);
+    void deleteByNotebookId(Long notebookId);
+}

core/src/main/java/com/loremind/infrastructure/persistence/jpa/RandomTableJpaRepository.java

@@ -0,0 +1,13 @@
+package com.loremind.infrastructure.persistence.jpa;
+
+import com.loremind.infrastructure.persistence.entity.RandomTableJpaEntity;
+import org.springframework.data.jpa.repository.JpaRepository;
+import org.springframework.stereotype.Repository;
+
+import java.util.List;
+
+@Repository
+public interface RandomTableJpaRepository extends JpaRepository<RandomTableJpaEntity, Long> {
+
+    List<RandomTableJpaEntity> findByCampaignIdOrderByOrderAsc(Long campaignId);
+}

core/src/main/java/com/loremind/infrastructure/persistence/postgres/PostgresNotebookRepository.java

@@ -0,0 +1,155 @@
+package com.loremind.infrastructure.persistence.postgres;
+
+import com.loremind.domain.campaigncontext.Notebook;
+import com.loremind.domain.campaigncontext.NotebookMessage;
+import com.loremind.domain.campaigncontext.NotebookSource;
+import com.loremind.domain.campaigncontext.ports.NotebookRepository;
+import com.loremind.infrastructure.persistence.entity.NotebookJpaEntity;
+import com.loremind.infrastructure.persistence.entity.NotebookMessageJpaEntity;
+import com.loremind.infrastructure.persistence.entity.NotebookSourceJpaEntity;
+import com.loremind.infrastructure.persistence.jpa.NotebookJpaRepository;
+import com.loremind.infrastructure.persistence.jpa.NotebookMessageJpaRepository;
+import com.loremind.infrastructure.persistence.jpa.NotebookSourceJpaRepository;
+import org.springframework.stereotype.Repository;
+import org.springframework.transaction.annotation.Transactional;
+
+import java.util.List;
+import java.util.Optional;
+import java.util.stream.Collectors;
+
+@Repository
+public class PostgresNotebookRepository implements NotebookRepository {
+
+    private final NotebookJpaRepository notebookJpa;
+    private final NotebookSourceJpaRepository sourceJpa;
+    private final NotebookMessageJpaRepository messageJpa;
+
+    public PostgresNotebookRepository(
+            NotebookJpaRepository notebookJpa,
+            NotebookSourceJpaRepository sourceJpa,
+            NotebookMessageJpaRepository messageJpa) {
+        this.notebookJpa = notebookJpa;
+        this.sourceJpa = sourceJpa;
+        this.messageJpa = messageJpa;
+    }
+
+    // --- Notebook ---
+
+    @Override
+    public Notebook save(Notebook notebook) {
+        NotebookJpaEntity entity = notebook.getId() != null
+                ? notebookJpa.findById(Long.parseLong(notebook.getId())).orElseGet(NotebookJpaEntity::new)
+                : new NotebookJpaEntity();
+        entity.setName(notebook.getName());
+        entity.setCampaignId(Long.parseLong(notebook.getCampaignId()));
+        return toNotebook(notebookJpa.save(entity));
+    }
+
+    @Override
+    public Optional<Notebook> findById(String id) {
+        return notebookJpa.findById(Long.parseLong(id)).map(this::toNotebook);
+    }
+
+    @Override
+    public List<Notebook> findByCampaignId(String campaignId) {
+        return notebookJpa.findByCampaignIdOrderByUpdatedAtDesc(Long.parseLong(campaignId)).stream()
+                .map(this::toNotebook).collect(Collectors.toList());
+    }
+
+    @Override
+    @Transactional
+    public void deleteById(String id) {
+        Long nid = Long.parseLong(id);
+        messageJpa.deleteByNotebookId(nid);
+        sourceJpa.deleteByNotebookId(nid);
+        notebookJpa.deleteById(nid);
+    }
+
+    @Override
+    public boolean existsById(String id) {
+        return notebookJpa.existsById(Long.parseLong(id));
+    }
+
+    // --- Sources ---
+
+    @Override
+    public NotebookSource saveSource(NotebookSource source) {
+        NotebookSourceJpaEntity entity = source.getId() != null
+                ? sourceJpa.findById(Long.parseLong(source.getId())).orElseGet(NotebookSourceJpaEntity::new)
+                : new NotebookSourceJpaEntity();
+        entity.setNotebookId(Long.parseLong(source.getNotebookId()));
+        entity.setFilename(source.getFilename());
+        entity.setStatus(source.getStatus());
+        entity.setChunkCount(source.getChunkCount());
+        entity.setPageCount(source.getPageCount());
+        return toSource(sourceJpa.save(entity));
+    }
+
+    @Override
+    public Optional<NotebookSource> findSourceById(String id) {
+        return sourceJpa.findById(Long.parseLong(id)).map(this::toSource);
+    }
+
+    @Override
+    public List<NotebookSource> findSourcesByNotebookId(String notebookId) {
+        return sourceJpa.findByNotebookIdOrderByCreatedAtAsc(Long.parseLong(notebookId)).stream()
+                .map(this::toSource).collect(Collectors.toList());
+    }
+
+    @Override
+    public void deleteSourceById(String id) {
+        sourceJpa.deleteById(Long.parseLong(id));
+    }
+
+    // --- Messages ---
+
+    @Override
+    public NotebookMessage saveMessage(NotebookMessage message) {
+        NotebookMessageJpaEntity entity = NotebookMessageJpaEntity.builder()
+                .notebookId(Long.parseLong(message.getNotebookId()))
+                .role(message.getRole())
+                .content(message.getContent())
+                .build();
+        return toMessage(messageJpa.save(entity));
+    }
+
+    @Override
+    public List<NotebookMessage> findMessagesByNotebookId(String notebookId) {
+        return messageJpa.findByNotebookIdOrderByCreatedAtAsc(Long.parseLong(notebookId)).stream()
+                .map(this::toMessage).collect(Collectors.toList());
+    }
+
+    // --- Mapping ---
+
+    private Notebook toNotebook(NotebookJpaEntity e) {
+        return Notebook.builder()
+                .id(e.getId().toString())
+                .name(e.getName())
+                .campaignId(e.getCampaignId().toString())
+                .createdAt(e.getCreatedAt())
+                .updatedAt(e.getUpdatedAt())
+                .build();
+    }
+
+    private NotebookSource toSource(NotebookSourceJpaEntity e) {
+        return NotebookSource.builder()
+                .id(e.getId().toString())
+                .notebookId(e.getNotebookId().toString())
+                .filename(e.getFilename())
+                .status(e.getStatus())
+                .chunkCount(e.getChunkCount())
+                .pageCount(e.getPageCount())
+                .createdAt(e.getCreatedAt())
+                .build();
+    }
+
+    private NotebookMessage toMessage(NotebookMessageJpaEntity e) {
+        return NotebookMessage.builder()
+                .id(e.getId().toString())
+                .notebookId(e.getNotebookId().toString())
+                .role(e.getRole())
+                .content(e.getContent())
+                .createdAt(e.getCreatedAt())
+                .build();
+    }
+}

core/src/main/java/com/loremind/infrastructure/persistence/postgres/PostgresRandomTableRepository.java

@@ -0,0 +1,106 @@
+package com.loremind.infrastructure.persistence.postgres;
+
+import com.loremind.domain.campaigncontext.RandomTable;
+import com.loremind.domain.campaigncontext.RandomTableEntry;
+import com.loremind.domain.campaigncontext.ports.RandomTableRepository;
+import com.loremind.infrastructure.persistence.entity.RandomTableEntryJpaEntity;
+import com.loremind.infrastructure.persistence.entity.RandomTableJpaEntity;
+import com.loremind.infrastructure.persistence.jpa.RandomTableJpaRepository;
+import org.springframework.stereotype.Repository;
+import org.springframework.transaction.annotation.Transactional;
+
+import java.util.ArrayList;
+import java.util.List;
+import java.util.Optional;
+import java.util.stream.Collectors;
+
+@Repository
+public class PostgresRandomTableRepository implements RandomTableRepository {
+
+    private final RandomTableJpaRepository jpaRepository;
+
+    public PostgresRandomTableRepository(RandomTableJpaRepository jpaRepository) {
+        this.jpaRepository = jpaRepository;
+    }
+
+    @Override
+    @Transactional
+    public RandomTable save(RandomTable table) {
+        // Création OU mise à jour : on charge l'entité gérée si elle existe afin que
+        // le remplacement des entrées (clear + add) déclenche bien orphanRemoval.
+        RandomTableJpaEntity entity = (table.getId() != null)
+                ? jpaRepository.findById(Long.parseLong(table.getId())).orElseGet(RandomTableJpaEntity::new)
+                : new RandomTableJpaEntity();
+
+        entity.setName(table.getName());
+        entity.setDescription(table.getDescription());
+        entity.setDiceFormula(table.getDiceFormula());
+        entity.setIcon(table.getIcon());
+        entity.setCampaignId(Long.parseLong(table.getCampaignId()));
+        entity.setOrder(table.getOrder());
+
+        // Remplacement en bloc des entrées (les anciennes sont supprimées via orphanRemoval).
+        entity.getEntries().clear();
+        int position = 0;
+        for (RandomTableEntry e : table.getEntries()) {
+            entity.getEntries().add(RandomTableEntryJpaEntity.builder()
+                    .minRoll(e.getMinRoll())
+                    .maxRoll(e.getMaxRoll())
+                    .label(e.getLabel())
+                    .detail(e.getDetail())
+                    .position(position++)
+                    .randomTable(entity)
+                    .build());
+        }
+
+        RandomTableJpaEntity saved = jpaRepository.save(entity);
+        return toDomainEntity(saved);
+    }
+
+    @Override
+    @Transactional(readOnly = true)
+    public Optional<RandomTable> findById(String id) {
+        return jpaRepository.findById(Long.parseLong(id)).map(this::toDomainEntity);
+    }
+
+    @Override
+    @Transactional(readOnly = true)
+    public List<RandomTable> findByCampaignId(String campaignId) {
+        return jpaRepository.findByCampaignIdOrderByOrderAsc(Long.parseLong(campaignId)).stream()
+                .map(this::toDomainEntity)
+                .collect(Collectors.toList());
+    }
+
+    @Override
+    public void deleteById(String id) {
+        jpaRepository.deleteById(Long.parseLong(id));
+    }
+
+    @Override
+    public boolean existsById(String id) {
+        return jpaRepository.existsById(Long.parseLong(id));
+    }
+
+    private RandomTable toDomainEntity(RandomTableJpaEntity e) {
+        List<RandomTableEntry> entries = e.getEntries().stream()
+                .map(c -> RandomTableEntry.builder()
+                        .minRoll(c.getMinRoll())
+                        .maxRoll(c.getMaxRoll())
+                        .label(c.getLabel())
+                        .detail(c.getDetail())
+                        .build())
+                .collect(Collectors.toCollection(ArrayList::new));
+        return RandomTable.builder()
+                .id(e.getId().toString())
+                .name(e.getName())
+                .description(e.getDescription())
+                .diceFormula(e.getDiceFormula())
+                .icon(e.getIcon())
+                .campaignId(e.getCampaignId().toString())
+                .order(e.getOrder())
+                .entries(entries)
+                .createdAt(e.getCreatedAt())
+                .updatedAt(e.getUpdatedAt())
+                .build();
+    }
+}

core/src/main/java/com/loremind/infrastructure/web/controller/NotebookController.java

@@ -0,0 +1,206 @@
+package com.loremind.infrastructure.web.controller;
+
+import com.loremind.application.campaigncontext.NotebookService;
+import com.loremind.domain.campaigncontext.Notebook;
+import com.loremind.domain.campaigncontext.NotebookSource;
+import com.loremind.domain.campaigncontext.ports.NotebookChatStreamer;
+import com.loremind.domain.campaigncontext.ports.NotebookException;
+import org.springframework.beans.factory.annotation.Qualifier;
+import org.springframework.core.task.TaskExecutor;
+import org.springframework.http.HttpStatus;
+import org.springframework.http.MediaType;
+import org.springframework.http.ResponseEntity;
+import org.springframework.web.bind.annotation.*;
+import org.springframework.web.multipart.MultipartFile;
+import org.springframework.web.server.ResponseStatusException;
+import org.springframework.web.servlet.mvc.method.annotation.SseEmitter;
+
+import java.io.IOException;
+import java.util.LinkedHashMap;
+import java.util.List;
+import java.util.Map;
+
+/**
+ * REST Controller des notebooks (atelier RAG). CRUD + upload/indexation de sources
+ * + chat ancré streamé (SSE) qui persiste la conversation.
+ */
+@RestController
+@RequestMapping("/api/notebooks")
+public class NotebookController {
+
+    private static final long SSE_TIMEOUT_MS = 10 * 60 * 1000L;
+
+    private final NotebookService service;
+    private final NotebookChatStreamer chatStreamer;
+    private final TaskExecutor taskExecutor;
+
+    public NotebookController(
+            NotebookService service,
+            NotebookChatStreamer chatStreamer,
+            @Qualifier("applicationTaskExecutor") TaskExecutor taskExecutor) {
+        this.service = service;
+        this.chatStreamer = chatStreamer;
+        this.taskExecutor = taskExecutor;
+    }
+
+    // --- Notebooks ---
+
+    @PostMapping
+    public ResponseEntity<Notebook> create(@RequestBody CreateRequest req) {
+        return ResponseEntity.ok(service.createNotebook(req.campaignId(), req.name()));
+    }
+
+    @GetMapping("/campaign/{campaignId}")
+    public ResponseEntity<List<Notebook>> listByCampaign(@PathVariable String campaignId) {
+        return ResponseEntity.ok(service.getNotebooksByCampaign(campaignId));
+    }
+
+    @GetMapping("/{id}")
+    public ResponseEntity<Map<String, Object>> get(@PathVariable String id) {
+        Notebook nb = service.getNotebook(id)
+                .orElseThrow(() -> new ResponseStatusException(HttpStatus.NOT_FOUND, "Notebook introuvable"));
+        Map<String, Object> out = new LinkedHashMap<>();
+        out.put("id", nb.getId());
+        out.put("name", nb.getName());
+        out.put("campaignId", nb.getCampaignId());
+        out.put("sources", service.getSources(id));
+        out.put("messages", service.getMessages(id));
+        return ResponseEntity.ok(out);
+    }
+
+    @PutMapping("/{id}")
+    public ResponseEntity<Notebook> rename(@PathVariable String id, @RequestBody RenameRequest req) {
+        return ResponseEntity.ok(service.renameNotebook(id, req.name()));
+    }
+
+    @DeleteMapping("/{id}")
+    public ResponseEntity<Void> delete(@PathVariable String id) {
+        service.deleteNotebook(id);
+        return ResponseEntity.noContent().build();
+    }
+
+    // --- Sources ---
+
+    @PostMapping("/{id}/sources")
+    public ResponseEntity<NotebookSource> addSource(
+            @PathVariable String id,
+            @RequestParam("file") MultipartFile file) {
+        try {
+            byte[] bytes = file.getBytes();
+            NotebookSource source = service.addSource(id, file.getOriginalFilename(), bytes);
+            return ResponseEntity.ok(source);
+        } catch (IOException e) {
+            throw new ResponseStatusException(HttpStatus.BAD_REQUEST, "Fichier illisible", e);
+        } catch (NotebookException e) {
+            throw new ResponseStatusException(HttpStatus.BAD_GATEWAY, e.getMessage(), e);
+        }
+    }
+
+    @DeleteMapping("/sources/{sourceId}")
+    public ResponseEntity<Void> deleteSource(@PathVariable String sourceId) {
+        service.deleteSource(sourceId);
+        return ResponseEntity.noContent().build();
+    }
+
+    // --- Chat ancré streamé ---
+
+    @PostMapping(value = "/{id}/chat/stream", produces = MediaType.TEXT_EVENT_STREAM_VALUE)
+    public SseEmitter chatStream(@PathVariable String id, @RequestBody ChatRequest req) {
+        SseEmitter emitter = new SseEmitter(SSE_TIMEOUT_MS);
+        Notebook nb = service.getNotebook(id)
+                .orElseThrow(() -> new ResponseStatusException(HttpStatus.NOT_FOUND, "Notebook introuvable"));
+
+        String userMessage = req.message() == null ? "" : req.message().trim();
+        if (userMessage.isEmpty()) {
+            fail(emitter, new IllegalArgumentException("Message vide."));
+            return emitter;
+        }
+        // Persiste le message utilisateur AVANT le stream (l'historique inclura ce tour).
+        service.addMessage(id, "user", userMessage);
+
+        List<NotebookChatStreamer.Msg> history = service.getMessages(id).stream()
+                .map(m -> new NotebookChatStreamer.Msg(m.getRole(), m.getContent()))
+                .toList();
+        List<String> sourceIds = service.readySourceIds(id);
+        String context = service.buildContext(nb.getCampaignId());
+
+        boolean deep = req.deep() != null && req.deep();
+        taskExecutor.execute(() -> {
+            StringBuilder assistant = new StringBuilder();
+            chatStreamer.stream(
+                    sourceIds, history, context, deep,
+                    token -> { assistant.append(token); sendToken(emitter, token); },
+                    progress -> sendProgress(emitter, progress),
+                    () -> {
+                        // Persiste la réponse de l'assistant à la fin du stream.
+                        if (assistant.length() > 0) {
+                            service.addMessage(id, "assistant", assistant.toString());
+                        }
+                        complete(emitter);
+                    },
+                    error -> fail(emitter, error));
+        });
+        return emitter;
+    }
+
+    // --- Helpers SSE (mêmes conventions que AiChatController) ---
+
+    private void sendToken(SseEmitter emitter, String token) {
+        try {
+            emitter.send(SseEmitter.event().name("token").data("{\"token\":" + jsonEscape(token) + "}"));
+        } catch (IOException e) {
+            emitter.completeWithError(e);
+        }
+    }
+
+    private void sendProgress(SseEmitter emitter, NotebookChatStreamer.Progress p) {
+        try {
+            emitter.send(SseEmitter.event().name("progress")
+                    .data("{\"current\":" + p.current() + ",\"total\":" + p.total() + "}"));
+        } catch (IOException e) {
+            emitter.completeWithError(e);
+        }
+    }
+
+    private void complete(SseEmitter emitter) {
+        try {
+            emitter.send(SseEmitter.event().name("done").data("{}"));
+            emitter.complete();
+        } catch (IOException e) {
+            emitter.completeWithError(e);
+        }
+    }
+
+    private void fail(SseEmitter emitter, Throwable error) {
+        try {
+            String message = error.getMessage() != null ? error.getMessage() : error.getClass().getSimpleName();
+            emitter.send(SseEmitter.event().name("error").data("{\"message\":" + jsonEscape(message) + "}"));
+            emitter.complete();
+        } catch (IOException ioe) {
+            emitter.completeWithError(ioe);
+        }
+    }
+
+    private String jsonEscape(String raw) {
+        if (raw == null) return "\"\"";
+        StringBuilder sb = new StringBuilder(raw.length() + 2).append('"');
+        for (int i = 0; i < raw.length(); i++) {
+            char c = raw.charAt(i);
+            switch (c) {
+                case '"':  sb.append("\\\""); break;
+                case '\\': sb.append("\\\\"); break;
+                case '\n': sb.append("\\n"); break;
+                case '\r': sb.append("\\r"); break;
+                case '\t': sb.append("\\t"); break;
+                default:
+                    if (c < 0x20) sb.append(String.format("\\u%04x", (int) c));
+                    else sb.append(c);
+            }
+        }
+        return sb.append('"').toString();
+    }
+
+    public record CreateRequest(String campaignId, String name) {}
+    public record RenameRequest(String name) {}
+    public record ChatRequest(String message, Boolean deep) {}
+}

core/src/main/java/com/loremind/infrastructure/web/controller/RandomTableController.java

@@ -0,0 +1,100 @@
+package com.loremind.infrastructure.web.controller;
+
+import com.loremind.application.campaigncontext.RandomTableService;
+import com.loremind.domain.campaigncontext.RandomTable;
+import com.loremind.domain.campaigncontext.ports.RandomTableGenerationException;
+import com.loremind.infrastructure.web.dto.campaigncontext.RandomTableDTO;
+import com.loremind.infrastructure.web.mapper.RandomTableMapper;
+import org.springframework.http.HttpStatus;
+import org.springframework.http.ResponseEntity;
+import org.springframework.web.bind.annotation.*;
+import org.springframework.web.server.ResponseStatusException;
+
+import java.util.List;
+import java.util.Map;
+import java.util.stream.Collectors;
+
+@RestController
+@RequestMapping("/api/random-tables")
+public class RandomTableController {
+
+    private final RandomTableService service;
+    private final RandomTableMapper mapper;
+
+    public RandomTableController(RandomTableService service, RandomTableMapper mapper) {
+        this.service = service;
+        this.mapper = mapper;
+    }
+
+    @PostMapping
+    public ResponseEntity<RandomTableDTO> create(@RequestBody RandomTableDTO dto) {
+        RandomTable created = service.createTable(toData(dto, null));
+        return ResponseEntity.ok(mapper.toDTO(created));
+    }
+
+    @GetMapping("/{id}")
+    public ResponseEntity<RandomTableDTO> getById(@PathVariable String id) {
+        return service.getTableById(id)
+                .map(t -> ResponseEntity.ok(mapper.toDTO(t)))
+                .orElse(ResponseEntity.notFound().build());
+    }
+
+    @GetMapping("/campaign/{campaignId}")
+    public ResponseEntity<List<RandomTableDTO>> getByCampaign(@PathVariable String campaignId) {
+        List<RandomTableDTO> dtos = service.getTablesByCampaignId(campaignId).stream()
+                .map(mapper::toDTO)
+                .collect(Collectors.toList());
+        return ResponseEntity.ok(dtos);
+    }
+
+    @PutMapping("/{id}")
+    public ResponseEntity<RandomTableDTO> update(@PathVariable String id, @RequestBody RandomTableDTO dto) {
+        RandomTable updated = service.updateTable(id, toData(dto, dto.getOrder()));
+        return ResponseEntity.ok(mapper.toDTO(updated));
+    }
+
+    @DeleteMapping("/{id}")
+    public ResponseEntity<Void> delete(@PathVariable String id) {
+        service.deleteTable(id);
+        return ResponseEntity.noContent().build();
+    }
+
+    /** Génère une PROPOSITION de table via l'IA (non persistée) — l'UI préremplit le formulaire. */
+    @PostMapping("/generate")
+    public ResponseEntity<RandomTableDTO> generate(@RequestBody GenerateRequest req) {
+        try {
+            RandomTable proposal = service.generateProposal(req.campaignId(), req.description(), req.diceFormula());
+            return ResponseEntity.ok(mapper.toDTO(proposal));
+        } catch (RandomTableGenerationException e) {
+            throw new ResponseStatusException(HttpStatus.BAD_GATEWAY, e.getMessage(), e);
+        }
+    }
+
+    /** Improvisation IA d'un court récit sur un résultat tiré (utilisé en partie). */
+    @PostMapping("/improvise")
+    public ResponseEntity<Map<String, String>> improvise(@RequestBody ImproviseRequest req) {
+        try {
+            String narration = service.improviseRoll(
+                    req.campaignId(), req.tableName(), req.resultLabel(), req.resultDetail());
+            return ResponseEntity.ok(Map.of("narration", narration));
+        } catch (RandomTableGenerationException e) {
+            throw new ResponseStatusException(HttpStatus.BAD_GATEWAY, e.getMessage(), e);
+        }
+    }
+
+    public record GenerateRequest(String campaignId, String description, String diceFormula) {}
+
+    public record ImproviseRequest(String campaignId, String tableName, String resultLabel, String resultDetail) {}
+
+    private RandomTableService.TableData toData(RandomTableDTO dto, Integer order) {
+        return new RandomTableService.TableData(
+                dto.getName(),
+                dto.getDescription(),
+                dto.getDiceFormula(),
+                dto.getIcon(),
+                mapper.toDomainEntries(dto.getEntries()),
+                dto.getCampaignId(),
+                order
+        );
+    }
+}

core/src/main/java/com/loremind/infrastructure/web/controller/SettingsController.java

@@ -140,6 +140,21 @@ public class SettingsController {
         return forward(HttpMethod.GET, "/models/onemin", null);
     }
 
+    @GetMapping("/models/openrouter")
+    public ResponseEntity<Map<String, Object>> listOpenRouterModels() {
+        return forward(HttpMethod.GET, "/models/openrouter", null);
+    }
+
+    @GetMapping("/models/mistral")
+    public ResponseEntity<Map<String, Object>> listMistralModels() {
+        return forward(HttpMethod.GET, "/models/mistral", null);
+    }
+
+    @GetMapping("/models/gemini")
+    public ResponseEntity<Map<String, Object>> listGeminiModels() {
+        return forward(HttpMethod.GET, "/models/gemini", null);
+    }
+
     /**
      * Serialiseur JSON minimal pour eviter d'instancier ObjectMapper a chaque
      * appel. Suffisant pour notre cas d'usage : Map<String,Object> avec des

core/src/main/java/com/loremind/infrastructure/web/dto/campaigncontext/RandomTableDTO.java

@@ -0,0 +1,21 @@
+package com.loremind.infrastructure.web.dto.campaigncontext;
+
+import lombok.Data;
+
+import java.util.ArrayList;
+import java.util.List;
+
+/**
+ * DTO d'une table aléatoire (avec ses entrées).
+ */
+@Data
+public class RandomTableDTO {
+    private String id;
+    private String name;
+    private String description;
+    private String diceFormula;
+    private String icon;
+    private String campaignId;
+    private int order;
+    private List<RandomTableEntryDTO> entries = new ArrayList<>();
+}

core/src/main/java/com/loremind/infrastructure/web/dto/campaigncontext/RandomTableEntryDTO.java

@@ -0,0 +1,14 @@
+package com.loremind.infrastructure.web.dto.campaigncontext;
+
+import lombok.Data;
+
+/**
+ * DTO d'une entrée de table aléatoire (plage de jet → résultat).
+ */
+@Data
+public class RandomTableEntryDTO {
+    private int minRoll;
+    private int maxRoll;
+    private String label;
+    private String detail;
+}

core/src/main/java/com/loremind/infrastructure/web/mapper/RandomTableMapper.java

@@ -0,0 +1,51 @@
+package com.loremind.infrastructure.web.mapper;
+
+import com.loremind.domain.campaigncontext.RandomTable;
+import com.loremind.domain.campaigncontext.RandomTableEntry;
+import com.loremind.infrastructure.web.dto.campaigncontext.RandomTableDTO;
+import com.loremind.infrastructure.web.dto.campaigncontext.RandomTableEntryDTO;
+import org.springframework.stereotype.Component;
+
+import java.util.List;
+import java.util.stream.Collectors;
+
+@Component
+public class RandomTableMapper {
+
+    public RandomTableDTO toDTO(RandomTable t) {
+        if (t == null) return null;
+        RandomTableDTO dto = new RandomTableDTO();
+        dto.setId(t.getId());
+        dto.setName(t.getName());
+        dto.setDescription(t.getDescription());
+        dto.setDiceFormula(t.getDiceFormula());
+        dto.setIcon(t.getIcon());
+        dto.setCampaignId(t.getCampaignId());
+        dto.setOrder(t.getOrder());
+        dto.setEntries(t.getEntries().stream().map(this::toEntryDTO).collect(Collectors.toList()));
+        return dto;
+    }
+
+    public List<RandomTableEntry> toDomainEntries(List<RandomTableEntryDTO> dtos) {
+        if (dtos == null) return List.of();
+        return dtos.stream().map(this::toDomainEntry).collect(Collectors.toList());
+    }
+
+    private RandomTableEntryDTO toEntryDTO(RandomTableEntry e) {
+        RandomTableEntryDTO dto = new RandomTableEntryDTO();
+        dto.setMinRoll(e.getMinRoll());
+        dto.setMaxRoll(e.getMaxRoll());
+        dto.setLabel(e.getLabel());
+        dto.setDetail(e.getDetail());
+        return dto;
+    }
+
+    private RandomTableEntry toDomainEntry(RandomTableEntryDTO dto) {
+        return RandomTableEntry.builder()
+                .minRoll(dto.getMinRoll())
+                .maxRoll(dto.getMaxRoll())
+                .label(dto.getLabel())
+                .detail(dto.getDetail())
+                .build();
+    }
+}

core/src/test/resources/application.properties

@@ -7,8 +7,8 @@
 # Surchargables via variables d'environnement (CI) :
 #   SPRING_DATASOURCE_URL, SPRING_DATASOURCE_USERNAME, SPRING_DATASOURCE_PASSWORD
 spring.datasource.url=${SPRING_DATASOURCE_URL:jdbc:postgresql://localhost:5432/loremind_test}
-spring.datasource.username=${SPRING_DATASOURCE_USERNAME:ietm64}
+spring.datasource.username=${SPRING_DATASOURCE_USERNAME:loremind_test}
-spring.datasource.password=${SPRING_DATASOURCE_PASSWORD:M&Ipourlavie64}
+spring.datasource.password=${SPRING_DATASOURCE_PASSWORD:loremind_test}
 spring.datasource.driver-class-name=org.postgresql.Driver
 
 # Configuration JPA pour les tests : schema recree a chaque run, pas de logs SQL.

web/nginx.conf

@@ -5,7 +5,9 @@ server {
     root /usr/share/nginx/html;
     index index.html;
 
-    client_max_body_size 10M;
+    # Upload max : aligné sur la limite multipart du Core (64 Mo). Couvre les
+    # imports/adaptations de PDF (livres de règles/campagne illustrés) et les images.
+    client_max_body_size 64M;
 
     location /api/ {
         proxy_pass http://core:8080/api/;
@@ -15,9 +17,11 @@ server {
         proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
         proxy_set_header X-Forwarded-Proto $scheme;
 
+        # Streaming SSE (chat, import, adaptation) : pas de buffering, et timeouts
+        # longs car une génération sur gros PDF peut tarder avant le 1er octet.
         proxy_buffering off;
-        proxy_read_timeout 300s;
+        proxy_read_timeout 600s;
-        proxy_send_timeout 300s;
+        proxy_send_timeout 600s;
     }
 
     # index.html : toujours revalide. Empeche un navigateur qui a precedemment

web/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "loremind-web",
-  "version": "0.10.0-beta",
+  "version": "0.11.0-beta",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "loremind-web",
-      "version": "0.10.0-beta",
+      "version": "0.11.0-beta",
       "dependencies": {
         "@angular/animations": "^17.0.0",
         "@angular/common": "^17.0.0",

web/package.json

@@ -1,6 +1,6 @@
 {
   "name": "loremind-web",
-  "version": "0.10.0-beta",
+  "version": "0.11.0-beta",
   "description": "LoreMind Frontend - Angular",
   "scripts": {
     "ng": "ng",

web/src/app/app.routes.ts

@@ -17,7 +17,6 @@ export const routes: Routes = [
   { path: 'campaigns', loadComponent: () => import('./campaigns/campaigns.component').then(m => m.CampaignsComponent) },
   { path: 'campaigns/:id', loadComponent: () => import('./campaigns/campaign/campaign-detail/campaign-detail.component').then(m => m.CampaignDetailComponent) },
   { path: 'campaigns/:campaignId/import', loadComponent: () => import('./campaigns/campaign/campaign-import/campaign-import.component').then(m => m.CampaignImportComponent) },
-  { path: 'campaigns/:campaignId/adapt', loadComponent: () => import('./campaigns/campaign/campaign-adapt/campaign-adapt.component').then(m => m.CampaignAdaptComponent) },
   { path: 'campaigns/:campaignId/playthroughs/:playthroughId', loadComponent: () => import('./campaigns/playthrough/playthrough-detail/playthrough-detail.component').then(m => m.PlaythroughDetailComponent) },
   { path: 'campaigns/:campaignId/playthroughs/:playthroughId/flags', loadComponent: () => import('./campaigns/playthrough/playthrough-flags-page/playthrough-flags-page.component').then(m => m.PlaythroughFlagsPageComponent) },
   { path: 'campaigns/:campaignId/playthroughs/:playthroughId/characters/create', loadComponent: () => import('./campaigns/character/character-edit/character-edit.component').then(m => m.CharacterEditComponent) },
@@ -26,6 +25,11 @@ export const routes: Routes = [
   { path: 'campaigns/:campaignId/npcs/create', loadComponent: () => import('./campaigns/npc/npc-edit/npc-edit.component').then(m => m.NpcEditComponent) },
   { path: 'campaigns/:campaignId/npcs/:npcId/edit', loadComponent: () => import('./campaigns/npc/npc-edit/npc-edit.component').then(m => m.NpcEditComponent) },
   { path: 'campaigns/:campaignId/npcs/:npcId', loadComponent: () => import('./campaigns/npc/npc-view/npc-view.component').then(m => m.NpcViewComponent) },
+  { path: 'campaigns/:campaignId/notebooks', loadComponent: () => import('./campaigns/notebook/notebook-list/notebook-list.component').then(m => m.NotebookListComponent) },
+  { path: 'campaigns/:campaignId/notebooks/:notebookId', loadComponent: () => import('./campaigns/notebook/notebook-detail/notebook-detail.component').then(m => m.NotebookDetailComponent) },
+  { path: 'campaigns/:campaignId/random-tables/create', loadComponent: () => import('./campaigns/random-table/random-table-edit/random-table-edit.component').then(m => m.RandomTableEditComponent) },
+  { path: 'campaigns/:campaignId/random-tables/:tableId/edit', loadComponent: () => import('./campaigns/random-table/random-table-edit/random-table-edit.component').then(m => m.RandomTableEditComponent) },
+  { path: 'campaigns/:campaignId/random-tables/:tableId', loadComponent: () => import('./campaigns/random-table/random-table-view/random-table-view.component').then(m => m.RandomTableViewComponent) },
   { path: 'campaigns/:campaignId/arcs/create', loadComponent: () => import('./campaigns/arc/arc-create/arc-create.component').then(m => m.ArcCreateComponent) },
   { path: 'campaigns/:campaignId/arcs/:arcId', loadComponent: () => import('./campaigns/arc/arc-view/arc-view.component').then(m => m.ArcViewComponent) },
   { path: 'campaigns/:campaignId/arcs/:arcId/edit', loadComponent: () => import('./campaigns/arc/arc-edit/arc-edit.component').then(m => m.ArcEditComponent) },

web/src/app/campaigns/arc/arc-create/arc-create.component.ts

@@ -7,6 +7,7 @@ import { LucideAngularModule, BookOpen } from 'lucide-angular';
 import { CampaignService } from '../../../services/campaign.service';
 import { CharacterService } from '../../../services/character.service';
 import { NpcService } from '../../../services/npc.service';
+import { RandomTableService } from '../../../services/random-table.service';
 import { LayoutService } from '../../../services/layout.service';
 import { loadCampaignTreeData, buildCampaignSidebarConfig } from '../../campaign-tree.helper';
 import { IconPickerComponent } from '../../../shared/icon-picker/icon-picker.component';
@@ -40,6 +41,7 @@ export class ArcCreateComponent implements OnInit, OnDestroy {
     private campaignService: CampaignService,
     private characterService: CharacterService,
     private npcService: NpcService,
+    private randomTableService: RandomTableService,
     private layoutService: LayoutService
   ) {
     this.form = this.fb.group({
@@ -59,7 +61,7 @@ export class ArcCreateComponent implements OnInit, OnDestroy {
     forkJoin({
       campaign: this.campaignService.getCampaignById(this.campaignId),
       allCampaigns: this.campaignService.getAllCampaigns(),
-      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService)
+      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService, this.randomTableService)
     }).subscribe(({ campaign, allCampaigns, treeData }) => {
       this.existingArcCount = treeData.arcs.length;
 

web/src/app/campaigns/arc/arc-edit/arc-edit.component.ts

@@ -8,6 +8,7 @@ import { LucideAngularModule, Trash2, Sparkles } from 'lucide-angular';
 import { CampaignService } from '../../../services/campaign.service';
 import { CharacterService } from '../../../services/character.service';
 import { NpcService } from '../../../services/npc.service';
+import { RandomTableService } from '../../../services/random-table.service';
 import { PageService } from '../../../services/page.service';
 import { LayoutService } from '../../../services/layout.service';
 import { PageTitleService } from '../../../services/page-title.service';
@@ -77,6 +78,7 @@ export class ArcEditComponent implements OnInit, OnDestroy {
     private campaignService: CampaignService,
     private characterService: CharacterService,
     private npcService: NpcService,
+    private randomTableService: RandomTableService,
     private pageService: PageService,
     private layoutService: LayoutService,
     private pageTitleService: PageTitleService,
@@ -116,7 +118,7 @@ export class ArcEditComponent implements OnInit, OnDestroy {
       campaign: this.campaignService.getCampaignById(this.campaignId),
       allCampaigns: this.campaignService.getAllCampaigns(),
       arc: this.campaignService.getArcById(this.arcId),
-      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService)
+      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService, this.randomTableService)
     }).pipe(
       switchMap(data => {
         const lid = data.campaign.loreId ?? null;

web/src/app/campaigns/arc/arc-view/arc-view.component.ts

@@ -8,6 +8,7 @@ import { resolveCampaignIcon } from '../../campaign-icons';
 import { CampaignService } from '../../../services/campaign.service';
 import { CharacterService } from '../../../services/character.service';
 import { NpcService } from '../../../services/npc.service';
+import { RandomTableService } from '../../../services/random-table.service';
 import { PageService } from '../../../services/page.service';
 import { LayoutService } from '../../../services/layout.service';
 import { PageTitleService } from '../../../services/page-title.service';
@@ -59,6 +60,7 @@ export class ArcViewComponent implements OnInit, OnDestroy {
     private campaignService: CampaignService,
     private characterService: CharacterService,
     private npcService: NpcService,
+    private randomTableService: RandomTableService,
     private pageService: PageService,
     private layoutService: LayoutService,
     private pageTitleService: PageTitleService,
@@ -82,7 +84,7 @@ export class ArcViewComponent implements OnInit, OnDestroy {
       campaign: this.campaignService.getCampaignById(this.campaignId),
       allCampaigns: this.campaignService.getAllCampaigns(),
       arc: this.campaignService.getArcById(this.arcId),
-      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService)
+      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService, this.randomTableService)
     }).pipe(
       switchMap(data => {
         const lid = data.campaign.loreId ?? null;

web/src/app/campaigns/campaign-tree.helper.ts

@@ -3,10 +3,13 @@ import { switchMap, map } from 'rxjs/operators';
 import { CampaignService } from '../services/campaign.service';
 import { CharacterService } from '../services/character.service';
 import { NpcService } from '../services/npc.service';
+import { RandomTableService } from '../services/random-table.service';
 import { TreeItem, SecondarySidebarConfig, GlobalItem } from '../services/layout.service';
 import { Arc, Chapter, Scene, Campaign } from '../services/campaign.model';
 import { Character } from '../services/character.model';
 import { Npc } from '../services/npc.model';
+import { RandomTable } from '../services/random-table.model';
+import { catchError } from 'rxjs/operators';
 
 /**
  * Helper — charge l'arborescence complète d'une campagne (arcs -> chapitres -> scènes)
@@ -22,13 +25,17 @@ export interface CampaignTreeData {
   scenesByChapter: Record<string, Scene[]>;
   characters: Character[];
   npcs: Npc[];
+  randomTables: RandomTable[];
 }
 
 export function loadCampaignTreeData(
   service: CampaignService,
   campaignId: string,
   characterService: CharacterService,
-  npcService: NpcService
+  npcService: NpcService,
+  // Optionnel pour ne pas casser les ~15 appelants existants : si fourni, les
+  // tables aléatoires sont chargées et apparaissent dans la sidebar.
+  randomTableService?: RandomTableService
 ): Observable<CampaignTreeData> {
   // Note refonte Playthrough : les PJ appartiennent désormais à une Partie,
   // pas à la campagne — on ne les charge plus ici (les vues qui les affichent
@@ -36,11 +43,14 @@ export function loadCampaignTreeData(
   return forkJoin({
     arcs: service.getArcs(campaignId),
     characters: of([] as Character[]),
-    npcs: npcService.getByCampaign(campaignId)
+    npcs: npcService.getByCampaign(campaignId),
+    randomTables: randomTableService
+      ? randomTableService.getByCampaign(campaignId).pipe(catchError(() => of([] as RandomTable[])))
+      : of([] as RandomTable[])
   }).pipe(
-    switchMap(({ arcs, characters, npcs }) => {
+    switchMap(({ arcs, characters, npcs, randomTables }) => {
       if (arcs.length === 0) {
-        return of({ arcs, chaptersByArc: {}, scenesByChapter: {}, characters, npcs });
+        return of({ arcs, chaptersByArc: {}, scenesByChapter: {}, characters, npcs, randomTables });
       }
       const chapterCalls = arcs.map(a =>
         service.getChapters(a.id!).pipe(map(chapters => ({ arcId: a.id!, chapters })))
@@ -55,7 +65,7 @@ export function loadCampaignTreeData(
           });
 
           if (allChapters.length === 0) {
-            return of({ arcs, chaptersByArc, scenesByChapter: {}, characters, npcs });
+            return of({ arcs, chaptersByArc, scenesByChapter: {}, characters, npcs, randomTables });
           }
           const sceneCalls = allChapters.map(c =>
             service.getScenes(c.id!).pipe(map(scenes => ({ chapterId: c.id!, scenes })))
@@ -64,7 +74,7 @@ export function loadCampaignTreeData(
             map(sceneResults => {
               const scenesByChapter: Record<string, Scene[]> = {};
               sceneResults.forEach(r => { scenesByChapter[r.chapterId] = r.scenes; });
-              return { arcs, chaptersByArc, scenesByChapter, characters, npcs };
+              return { arcs, chaptersByArc, scenesByChapter, characters, npcs, randomTables };
             })
           );
         })
@@ -127,6 +137,8 @@ export function buildCampaignTree(campaignId: string, data: CampaignTreeData): T
         id: `chapter-${ch.id}`,
         label: ch.name,
         iconKey: ch.icon ?? undefined,
+        // Cadenas si le chapitre porte des conditions de déblocage (hub ou linéaire).
+        meta: (ch.prerequisites?.length ?? 0) > 0 ? '🔒' : undefined,
         children: sceneItems,
         route: `/campaigns/${campaignId}/arcs/${arc.id}/chapters/${ch.id}`,
         createActions: [{
@@ -155,7 +167,46 @@ export function buildCampaignTree(campaignId: string, data: CampaignTreeData): T
     };
   });
 
-  return [...arcNodes, npcsNode];
+  const sortedTables = [...(data.randomTables ?? [])].sort(byName);
+  const tableItems: TreeItem[] = sortedTables.map(t => ({
+    id: `random-table-${t.id}`,
+    label: t.name,
+    iconKey: t.icon ?? 'dice',
+    route: `/campaigns/${campaignId}/random-tables/${t.id}`
+  }));
+
+  const tablesNode: TreeItem = {
+    id: 'random-tables-root',
+    label: 'Tables aléatoires',
+    iconKey: 'dice',
+    children: tableItems,
+    meta: tableItems.length ? String(tableItems.length) : undefined,
+    sectionHeaderBefore: 'Outils',
+    createActions: [{
+      id: 'new-random-table',
+      label: 'Nouvelle table',
+      route: `/campaigns/${campaignId}/random-tables/create`,
+      actionIcon: 'plus'
+    }]
+  };
+
+  // Lien simple vers les ateliers (la liste se charge sur sa page — pas de fetch ici).
+  const notebooksNode: TreeItem = {
+    id: 'notebooks-root',
+    label: 'Ateliers (IA + PDF)',
+    iconKey: 'book-open',
+    route: `/campaigns/${campaignId}/notebooks`
+  };
+
+  // Importer un PDF de campagne → arborescence (outil, comme tables & ateliers).
+  const importNode: TreeItem = {
+    id: 'import-pdf-root',
+    label: 'Importer un PDF',
+    iconKey: 'file-up',
+    route: `/campaigns/${campaignId}/import`
+  };
+
+  return [...arcNodes, npcsNode, tablesNode, notebooksNode, importNode];
 }
 
 /**

web/src/app/campaigns/campaign/campaign-adapt/campaign-adapt.component.html

@@ -1,61 +0,0 @@
-<div class="adapt-page">
-
-  <div class="page-header">
-    <button type="button" class="btn-back" (click)="back()">
-      <lucide-icon [img]="ArrowLeft" [size]="14"></lucide-icon>
-      Retour à la campagne
-    </button>
-    <h1>Adapter un PDF à cette campagne</h1>
-    <p class="subtitle">
-      L'IA connaît votre campagne (structure, PNJ, univers) et lit le PDF, puis discute avec vous
-      pour l'intégrer et l'adapter. Ce sont des <strong>conseils</strong> : rien n'est créé,
-      vous appliquez ce qui vous plaît — et vous pouvez lui répondre pour qu'elle ajuste.
-    </p>
-  </div>
-
-  <!-- Choix / changement du PDF -->
-  <section class="upload-bar">
-    <input #pdfInput type="file" accept="application/pdf,.pdf" hidden (change)="onPdfSelected($event)" />
-    <button type="button" class="btn-primary" [disabled]="streaming" (click)="pdfInput.click()">
-      <lucide-icon [img]="Upload" [size]="15"></lucide-icon>
-      {{ hasConversation ? 'Changer de PDF' : 'Choisir un PDF à adapter' }}
-    </button>
-    <span class="file-name" *ngIf="fileName">{{ fileName }}</span>
-  </section>
-
-  <p class="adapt-error" *ngIf="error">{{ error }}</p>
-
-  <!-- Conversation -->
-  <section class="chat" *ngIf="hasConversation">
-    <div class="msg" *ngFor="let m of messages; let i = index"
-         [class.msg--user]="m.role === 'user'" [class.msg--assistant]="m.role === 'assistant'">
-      <div class="msg-role">{{ m.role === 'user' ? 'Vous' : 'IA' }}</div>
-
-      <div class="msg-body" *ngIf="m.role === 'user'">{{ m.content }}</div>
-      <div class="msg-body markdown-body" *ngIf="m.role === 'assistant'" [innerHTML]="m.content | markdown"></div>
-
-      <div class="msg-actions" *ngIf="m.role === 'assistant' && m.content && !(streaming && i === messages.length - 1)">
-        <button type="button" class="btn-copy" (click)="copy(i)">
-          <lucide-icon [img]="copiedIndex === i ? Check : Copy" [size]="12"></lucide-icon>
-          {{ copiedIndex === i ? 'Copié' : 'Copier' }}
-        </button>
-      </div>
-
-      <span class="streaming-cursor" *ngIf="streaming && i === messages.length - 1 && m.role === 'assistant'">▌</span>
-    </div>
-  </section>
-
-  <!-- Saisie d'un feedback -->
-  <section class="composer" *ngIf="hasConversation">
-    <textarea
-      [(ngModel)]="input"
-      [disabled]="streaming"
-      rows="2"
-      placeholder="Répondez à l'IA : corrigez, demandez une alternative, un autre lieu à intégrer…"
-      (keydown.enter)="$event.preventDefault(); sendCurrent()"></textarea>
-    <button type="button" class="btn-send" [disabled]="streaming || !input.trim()" (click)="sendCurrent()">
-      <lucide-icon [img]="Send" [size]="16"></lucide-icon>
-    </button>
-  </section>
-
-</div>

web/src/app/campaigns/campaign/campaign-adapt/campaign-adapt.component.scss

@@ -1,184 +0,0 @@
-.adapt-page {
-  padding: 2rem 2.5rem;
-  max-width: 900px;
-}
-
-.page-header {
-  margin-bottom: 1.25rem;
-
-  h1 { margin: 0.6rem 0 0.3rem; font-size: 1.5rem; color: #f3f4f6; }
-}
-
-.btn-back {
-  display: inline-flex;
-  align-items: center;
-  gap: 0.4rem;
-  background: transparent;
-  border: none;
-  color: #9ca3af;
-  cursor: pointer;
-  font-size: 0.85rem;
-  padding: 0;
-
-  &:hover { color: #c4b5fd; }
-}
-
-.subtitle { margin: 0; color: #9ca3af; font-size: 0.9rem; strong { color: #d1d5db; } }
-
-// --- Barre PDF --------------------------------------------------------------
-.upload-bar {
-  display: flex;
-  align-items: center;
-  gap: 0.8rem;
-  flex-wrap: wrap;
-  margin-bottom: 1rem;
-}
-
-.btn-primary {
-  display: inline-flex;
-  align-items: center;
-  gap: 0.45rem;
-  padding: 0.6rem 1.1rem;
-  border-radius: 8px;
-  font-size: 0.9rem;
-  font-weight: 500;
-  cursor: pointer;
-  border: none;
-  background: #6c63ff;
-  color: white;
-
-  &:hover:not(:disabled) { background: #5b52e0; }
-  &:disabled { opacity: 0.6; cursor: progress; }
-}
-
-.file-name { color: #9ca3af; font-size: 0.82rem; }
-
-.adapt-error {
-  margin: 0 0 1rem;
-  padding: 0.55rem 0.8rem;
-  background: rgba(248, 113, 113, 0.1);
-  border: 1px solid rgba(248, 113, 113, 0.35);
-  border-radius: 8px;
-  color: #fca5a5;
-  font-size: 0.85rem;
-}
-
-// --- Conversation -----------------------------------------------------------
-.chat {
-  display: flex;
-  flex-direction: column;
-  gap: 0.9rem;
-  margin-bottom: 1rem;
-}
-
-.msg {
-  border-radius: 12px;
-  padding: 0.7rem 0.9rem;
-  border: 1px solid #1f2937;
-
-  &--user {
-    background: rgba(108, 99, 255, 0.1);
-    border-color: rgba(108, 99, 255, 0.3);
-    align-self: flex-end;
-    max-width: 85%;
-  }
-
-  &--assistant {
-    background: #0b1220;
-  }
-}
-
-.msg-role {
-  font-size: 0.68rem;
-  text-transform: uppercase;
-  letter-spacing: 0.05em;
-  color: #9ca3af;
-  margin-bottom: 0.35rem;
-}
-
-.msg-body {
-  color: #d1d5db;
-  font-size: 0.92rem;
-  line-height: 1.6;
-  white-space: pre-wrap;
-
-  &.markdown-body { white-space: normal; }
-
-  ::ng-deep {
-    h1, h2, h3 { color: #f3f4f6; margin: 0.9rem 0 0.4rem; }
-    h2 { font-size: 1.08rem; }
-    h3 { font-size: 1rem; }
-    ul, ol { padding-left: 1.3rem; }
-    li { margin: 0.2rem 0; }
-    strong { color: #fff; }
-    code { background: #1f2937; padding: 0.1rem 0.3rem; border-radius: 4px; font-size: 0.85em; }
-    a { color: #a78bfa; }
-    p { margin: 0.4rem 0; }
-  }
-}
-
-.msg-actions { margin-top: 0.5rem; }
-
-.btn-copy {
-  display: inline-flex;
-  align-items: center;
-  gap: 0.3rem;
-  background: transparent;
-  border: 1px solid #374151;
-  border-radius: 6px;
-  color: #9ca3af;
-  cursor: pointer;
-  font-size: 0.75rem;
-  padding: 0.2rem 0.5rem;
-
-  &:hover { color: #c4b5fd; border-color: #6c63ff; }
-}
-
-.streaming-cursor {
-  display: inline-block;
-  color: #6c63ff;
-  animation: blink 1s step-start infinite;
-}
-
-@keyframes blink { 50% { opacity: 0; } }
-
-// --- Composer ---------------------------------------------------------------
-.composer {
-  display: flex;
-  gap: 0.5rem;
-  align-items: flex-end;
-  position: sticky;
-  bottom: 0;
-  padding-top: 0.5rem;
-  background: linear-gradient(to top, var(--color-bg, #0a0f1a) 70%, transparent);
-
-  textarea {
-    flex: 1;
-    background: #0b1220;
-    border: 1px solid #1f2937;
-    border-radius: 8px;
-    color: #f3f4f6;
-    padding: 0.55rem 0.7rem;
-    font-size: 0.9rem;
-    font-family: inherit;
-    resize: vertical;
-    &:focus { outline: none; border-color: #6c63ff; }
-    &:disabled { opacity: 0.6; }
-  }
-}
-
-.btn-send {
-  display: inline-flex;
-  align-items: center;
-  justify-content: center;
-  width: 42px;
-  height: 42px;
-  border-radius: 8px;
-  border: none;
-  background: #6c63ff;
-  color: white;
-  cursor: pointer;
-
-  &:hover:not(:disabled) { background: #5b52e0; }
-  &:disabled { opacity: 0.5; cursor: default; }
-}

web/src/app/campaigns/campaign/campaign-adapt/campaign-adapt.component.ts

@@ -1,130 +0,0 @@
-import { Component, OnInit } from '@angular/core';
-import { CommonModule } from '@angular/common';
-import { ActivatedRoute, Router } from '@angular/router';
-import { FormsModule } from '@angular/forms';
-import { LucideAngularModule, ArrowLeft, Upload, Copy, Check, Send } from 'lucide-angular';
-import { CampaignAdaptService, AdaptMessage } from '../../../services/campaign-adapt.service';
-import { CampaignSidebarService } from '../../../services/campaign-sidebar.service';
-import { PageTitleService } from '../../../services/page-title.service';
-import { MarkdownPipe } from '../../../shared/markdown.pipe';
-
-const FIRST_PROMPT = 'Propose-moi comment intégrer et adapter ce PDF à ma campagne.';
-
-/**
- * Page « Adapter un PDF » — CONVERSATIONNELLE. L'IA connaît la campagne (structure,
- * PNJ, univers) + lit le PDF, propose une 1re adaptation, puis l'utilisateur peut
- * répondre (corriger, demander des alternatives…) et l'IA rebondit.
- * Route : /campaigns/:campaignId/adapt — rien n'est créé, conseils à appliquer à la main.
- */
-@Component({
-  selector: 'app-campaign-adapt',
-  standalone: true,
-  imports: [CommonModule, FormsModule, LucideAngularModule, MarkdownPipe],
-  templateUrl: './campaign-adapt.component.html',
-  styleUrls: ['./campaign-adapt.component.scss']
-})
-export class CampaignAdaptComponent implements OnInit {
-  readonly ArrowLeft = ArrowLeft;
-  readonly Upload = Upload;
-  readonly Copy = Copy;
-  readonly Check = Check;
-  readonly Send = Send;
-
-  campaignId = '';
-
-  /** PDF choisi, conservé pour les tours de conversation suivants. */
-  private file: File | null = null;
-  fileName = '';
-
-  /** Conversation affichée (user + assistant). */
-  messages: AdaptMessage[] = [];
-  streaming = false;
-  error: string | null = null;
-
-  /** Saisie du message en cours. */
-  input = '';
-  copiedIndex: number | null = null;
-
-  constructor(
-    private route: ActivatedRoute,
-    private router: Router,
-    private service: CampaignAdaptService,
-    private campaignSidebar: CampaignSidebarService,
-    private pageTitle: PageTitleService
-  ) {}
-
-  ngOnInit(): void {
-    this.campaignId = this.route.snapshot.paramMap.get('campaignId')!;
-    this.pageTitle.set('Adapter un PDF');
-    this.campaignSidebar.show(this.campaignId);
-  }
-
-  get hasConversation(): boolean { return this.messages.length > 0; }
-
-  // --- Choix du PDF (démarre / réinitialise la conversation) ---------------
-
-  onPdfSelected(event: Event): void {
-    const input = event.target as HTMLInputElement;
-    const file = input.files?.[0];
-    input.value = '';
-    if (!file) return;
-    this.file = file;
-    this.fileName = file.name;
-    this.messages = [];
-    this.error = null;
-    this.send(FIRST_PROMPT);
-  }
-
-  // --- Envoi d'un message (1er tour ou feedback) ---------------------------
-
-  sendCurrent(): void {
-    const text = this.input.trim();
-    if (!text || this.streaming) return;
-    this.input = '';
-    this.send(text);
-  }
-
-  private send(text: string): void {
-    if (!this.file || this.streaming) return;
-    this.error = null;
-
-    this.messages.push({ role: 'user', content: text });
-    // Historique envoyé = tout jusqu'au message user inclus (sans la bulle vide).
-    const payload: AdaptMessage[] = this.messages.map(m => ({ role: m.role, content: m.content }));
-
-    const assistant: AdaptMessage = { role: 'assistant', content: '' };
-    this.messages.push(assistant);
-    this.streaming = true;
-
-    this.service.adviseStream(this.campaignId, this.file, payload).subscribe({
-      next: (ev) => {
-        if (ev.type === 'token') {
-          assistant.content += ev.value;
-        } else if (ev.type === 'done') {
-          this.streaming = false;
-        }
-      },
-      error: (err: Error) => {
-        this.streaming = false;
-        // Bulle assistant restée vide → on la retire pour ne pas afficher de vide.
-        if (!assistant.content) {
-          this.messages = this.messages.filter(m => m !== assistant);
-        }
-        this.error = err?.message ? `Échec : ${err.message}` : "Échec de l'adaptation.";
-      }
-    });
-  }
-
-  copy(index: number): void {
-    const msg = this.messages[index];
-    if (!msg) return;
-    navigator.clipboard?.writeText(msg.content).then(() => {
-      this.copiedIndex = index;
-      setTimeout(() => (this.copiedIndex = null), 2000);
-    });
-  }
-
-  back(): void {
-    this.router.navigate(['/campaigns', this.campaignId]);
-  }
-}

web/src/app/campaigns/campaign/campaign-detail/campaign-detail.component.html

@@ -144,14 +144,6 @@
     <div class="section-header">
       <h2>Arcs narratifs</h2>
       <div class="section-header-actions">
-        <button class="btn-add btn-add--secondary" (click)="adaptCampaign()" title="Conseils IA pour adapter un PDF à cette campagne">
-          <lucide-icon [img]="Sparkles" [size]="14"></lucide-icon>
-          Adapter un PDF
-        </button>
-        <button class="btn-add btn-add--secondary" (click)="importCampaign()" title="Générer l'arborescence depuis un PDF de campagne">
-          <lucide-icon [img]="Upload" [size]="14"></lucide-icon>
-          Importer un PDF
-        </button>
         <button class="btn-add" (click)="createArc()">
           <lucide-icon [img]="Plus" [size]="14"></lucide-icon>
           Nouvel arc

web/src/app/campaigns/campaign/campaign-detail/campaign-detail.component.ts

@@ -12,6 +12,7 @@ import { GameSystemService } from '../../../services/game-system.service';
 import { GameSystem } from '../../../services/game-system.model';
 import { CharacterService } from '../../../services/character.service';
 import { NpcService } from '../../../services/npc.service';
+import { RandomTableService } from '../../../services/random-table.service';
 import { SessionService } from '../../../services/session.service';
 import { PlaythroughService } from '../../../services/playthrough.service';
 import { Playthrough } from '../../../services/campaign.model';
@@ -94,6 +95,7 @@ export class CampaignDetailComponent implements OnInit, OnDestroy {
     private gameSystemService: GameSystemService,
     private characterService: CharacterService,
     private npcService: NpcService,
+    private randomTableService: RandomTableService,
     private sessionService: SessionService,
     private playthroughService: PlaythroughService,
     private layoutService: LayoutService,
@@ -111,8 +113,8 @@ export class CampaignDetailComponent implements OnInit, OnDestroy {
       switchMap(id => forkJoin({
         campaign: this.campaignService.getCampaignById(id),
         allCampaigns: this.campaignService.getAllCampaigns(),
-        treeData: loadCampaignTreeData(this.campaignService, id, this.characterService, this.npcService).pipe(
+        treeData: loadCampaignTreeData(this.campaignService, id, this.characterService, this.npcService, this.randomTableService).pipe(
-          catchError(() => of({ arcs: [], chaptersByArc: {}, scenesByChapter: {}, characters: [], npcs: [] } as CampaignTreeData))
+          catchError(() => of({ arcs: [], chaptersByArc: {}, scenesByChapter: {}, characters: [], npcs: [], randomTables: [] } as CampaignTreeData))
         ),
         playthroughs: this.playthroughService.listByCampaign(id).pipe(catchError(() => of([] as Playthrough[])))
       }))
@@ -148,8 +150,8 @@ export class CampaignDetailComponent implements OnInit, OnDestroy {
     forkJoin({
       campaign: this.campaignService.getCampaignById(id),
       allCampaigns: this.campaignService.getAllCampaigns(),
-      treeData: loadCampaignTreeData(this.campaignService, id, this.characterService, this.npcService).pipe(
+      treeData: loadCampaignTreeData(this.campaignService, id, this.characterService, this.npcService, this.randomTableService).pipe(
-        catchError(() => of({ arcs: [], chaptersByArc: {}, scenesByChapter: {}, characters: [], npcs: [] } as CampaignTreeData))
+        catchError(() => of({ arcs: [], chaptersByArc: {}, scenesByChapter: {}, characters: [], npcs: [], randomTables: [] } as CampaignTreeData))
       ),
       playthroughs: this.playthroughService.listByCampaign(id).pipe(catchError(() => of([] as Playthrough[])))
     }).subscribe(({ campaign, allCampaigns, treeData, playthroughs }) => {
@@ -246,18 +248,6 @@ export class CampaignDetailComponent implements OnInit, OnDestroy {
     this.router.navigate(['/campaigns', this.campaign.id, 'arcs', 'create']);
   }
 
-  /** Ouvre la page d'import d'un PDF de campagne (proposition d'arbre à réviser). */
-  importCampaign(): void {
-    if (!this.campaign) return;
-    this.router.navigate(['/campaigns', this.campaign.id, 'import']);
-  }
-
-  /** Ouvre la page de conseils d'adaptation d'un PDF à cette campagne. */
-  adaptCampaign(): void {
-    if (!this.campaign) return;
-    this.router.navigate(['/campaigns', this.campaign.id, 'adapt']);
-  }
-
   openArc(arc: Arc): void {
     if (!this.campaign || !arc.id) return;
     this.router.navigate(['/campaigns', this.campaign.id, 'arcs', arc.id]);

web/src/app/campaigns/campaign/campaign-import/campaign-import.component.ts

@@ -10,6 +10,7 @@ import { CampaignImportService } from '../../../services/campaign-import.service
 import { CampaignService } from '../../../services/campaign.service';
 import { CharacterService } from '../../../services/character.service';
 import { NpcService } from '../../../services/npc.service';
+import { RandomTableService } from '../../../services/random-table.service';
 import { CampaignSidebarService } from '../../../services/campaign-sidebar.service';
 import { PageTitleService } from '../../../services/page-title.service';
 import { ArcKind, ArcProposal, ChapterProposal, SceneProposal } from '../../../services/campaign-import.model';
@@ -91,6 +92,7 @@ export class CampaignImportComponent implements OnInit {
     private campaignService: CampaignService,
     private characterService: CharacterService,
     private npcService: NpcService,
+    private randomTableService: RandomTableService,
     private campaignSidebar: CampaignSidebarService,
     private pageTitle: PageTitleService
   ) {}
@@ -102,7 +104,7 @@ export class CampaignImportComponent implements OnInit {
 
     // Pré-chargement de l'arborescence existante (pour fusionner à la revue).
     // En cas d'échec on dégrade : tout sera considéré comme nouveau.
-    loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService)
+    loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService, this.randomTableService)
       .pipe(catchError(() => of(null)))
       .subscribe(data => this.existingData = data);
   }

web/src/app/campaigns/chapter/chapter-create/chapter-create.component.ts

@@ -7,6 +7,7 @@ import { LucideAngularModule } from 'lucide-angular';
 import { CampaignService } from '../../../services/campaign.service';
 import { CharacterService } from '../../../services/character.service';
 import { NpcService } from '../../../services/npc.service';
+import { RandomTableService } from '../../../services/random-table.service';
 import { LayoutService } from '../../../services/layout.service';
 import { loadCampaignTreeData, buildCampaignSidebarConfig } from '../../campaign-tree.helper';
 import { IconPickerComponent } from '../../../shared/icon-picker/icon-picker.component';
@@ -42,6 +43,7 @@ export class ChapterCreateComponent implements OnInit, OnDestroy {
     private campaignService: CampaignService,
     private characterService: CharacterService,
     private npcService: NpcService,
+    private randomTableService: RandomTableService,
     private layoutService: LayoutService
   ) {
     this.form = this.fb.group({
@@ -60,7 +62,7 @@ export class ChapterCreateComponent implements OnInit, OnDestroy {
     forkJoin({
       campaign: this.campaignService.getCampaignById(this.campaignId),
       allCampaigns: this.campaignService.getAllCampaigns(),
-      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService)
+      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService, this.randomTableService)
     }).subscribe(({ campaign, allCampaigns, treeData }) => {
       const currentArc = treeData.arcs.find(a => a.id === this.arcId);
       this.arcName = currentArc?.name ?? '';

web/src/app/campaigns/chapter/chapter-edit/chapter-edit.component.html

@@ -26,13 +26,17 @@
 
   <form [formGroup]="form" (ngSubmit)="submit()" class="edit-form">
 
-    <!-- Section Hub : conditions de déblocage (scénario). Visible si l'arc parent est HUB. -->
+    <!-- Conditions de déblocage (scénario) — pour TOUT chapitre : arc HUB (= quête)
-    <div class="hub-section" *ngIf="parentArc?.type === 'HUB'">
+         comme arc linéaire (= chapitre conditionnel). Vide = disponible d'emblée. -->
-      <h2 class="hub-section-title">🗺️ Conditions de déblocage de la quête</h2>
+    <div class="hub-section">
+      <h2 class="hub-section-title">
+        {{ parentArc?.type === 'HUB' ? '🗺️ Conditions de déblocage de la quête' : '🔒 Conditions de déblocage du chapitre' }}
+      </h2>
       <small class="field-hint">
-        Définition du scénario : toutes les conditions doivent être remplies (ET) pour
+        Définition du scénario : toutes les conditions doivent être remplies (ET) pour que
-        que la quête soit débloquée dans une Partie. La progression en cours et l'état
+        {{ parentArc?.type === 'HUB' ? 'la quête' : 'ce chapitre' }} soit débloqué(e) dans une Partie.
-        des faits se gèrent dans l'écran de la Partie, pas ici.
+        Laissez vide pour qu'il soit disponible dès le départ. La progression et l'état des
+        faits se gèrent dans l'écran de la Partie, pas ici.
       </small>
       <app-prerequisite-editor
         [prerequisites]="prerequisites"

web/src/app/campaigns/chapter/chapter-edit/chapter-edit.component.ts

@@ -8,6 +8,7 @@ import { LucideAngularModule, Trash2, Sparkles } from 'lucide-angular';
 import { CampaignService } from '../../../services/campaign.service';
 import { CharacterService } from '../../../services/character.service';
 import { NpcService } from '../../../services/npc.service';
+import { RandomTableService } from '../../../services/random-table.service';
 import { PageService } from '../../../services/page.service';
 import { LayoutService } from '../../../services/layout.service';
 import { PageTitleService } from '../../../services/page-title.service';
@@ -92,6 +93,7 @@ export class ChapterEditComponent implements OnInit, OnDestroy {
     private campaignService: CampaignService,
     private characterService: CharacterService,
     private npcService: NpcService,
+    private randomTableService: RandomTableService,
     private pageService: PageService,
     private layoutService: LayoutService,
     private pageTitleService: PageTitleService,
@@ -128,7 +130,7 @@ export class ChapterEditComponent implements OnInit, OnDestroy {
       campaign: this.campaignService.getCampaignById(this.campaignId),
       allCampaigns: this.campaignService.getAllCampaigns(),
       chapter: this.campaignService.getChapterById(this.chapterId),
-      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService)
+      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService, this.randomTableService)
     }).pipe(
       switchMap(data => {
         const lid = data.campaign.loreId ?? null;

web/src/app/campaigns/chapter/chapter-graph/chapter-graph.component.ts

@@ -6,6 +6,7 @@ import { LucideAngularModule, ArrowLeft } from 'lucide-angular';
 import { CampaignService } from '../../../services/campaign.service';
 import { CharacterService } from '../../../services/character.service';
 import { NpcService } from '../../../services/npc.service';
+import { RandomTableService } from '../../../services/random-table.service';
 import { LayoutService, GlobalItem } from '../../../services/layout.service';
 import { PageTitleService } from '../../../services/page-title.service';
 import { Campaign, Chapter, Scene } from '../../../services/campaign.model';
@@ -70,6 +71,7 @@ export class ChapterGraphComponent implements OnInit, OnDestroy {
     private campaignService: CampaignService,
     private characterService: CharacterService,
     private npcService: NpcService,
+    private randomTableService: RandomTableService,
     private layoutService: LayoutService,
     private pageTitleService: PageTitleService
   ) {}
@@ -89,7 +91,7 @@ export class ChapterGraphComponent implements OnInit, OnDestroy {
       allCampaigns: this.campaignService.getAllCampaigns(),
       chapter: this.campaignService.getChapterById(this.chapterId),
       scenes: this.campaignService.getScenes(this.chapterId),
-      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService)
+      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService, this.randomTableService)
     }).subscribe(({ campaign, allCampaigns, chapter, scenes, treeData }) => {
       this.chapter = chapter;
       this.scenes = scenes;

web/src/app/campaigns/chapter/chapter-view/chapter-view.component.html

@@ -6,7 +6,14 @@
         <lucide-icon *ngIf="chapter.icon" [img]="resolveCampaignIcon(chapter.icon)" [size]="22" class="title-icon"></lucide-icon>
         {{ chapter.name }}
       </h1>
-      <p class="view-subtitle">{{ parentArc?.type === 'HUB' ? 'Quête (Hub)' : 'Chapitre' }}</p>
+      <p class="view-subtitle">
+        {{ parentArc?.type === 'HUB' ? 'Quête (Hub)' : 'Chapitre' }}
+        <span class="cond-badge" *ngIf="(chapter.prerequisites?.length ?? 0) > 0"
+              title="Ce chapitre a des conditions de déblocage : il se débloque en Partie quand elles sont remplies.">
+          <lucide-icon [img]="Lock" [size]="12"></lucide-icon>
+          Conditionnel
+        </span>
+      </p>
     </div>
     <div class="view-actions">
       <button type="button" class="btn-secondary" (click)="openGraph()"
@@ -25,8 +32,9 @@
     </div>
   </header>
 
-  <!-- Conditions de déblocage (donnée de scénario, read-only). Toujours visible si Arc HUB. -->
+  <!-- Conditions de déblocage (scénario, read-only). Visible pour un arc HUB (quête)
-  <section class="view-section" *ngIf="parentArc?.type === 'HUB'">
+       OU dès qu'un chapitre linéaire porte des conditions. -->
+  <section class="view-section" *ngIf="parentArc?.type === 'HUB' || (chapter.prerequisites?.length ?? 0) > 0">
     <h2 class="view-section-title"><span class="view-section-icon">🔒</span> Conditions de déblocage</h2>
 
     <ng-container *ngIf="(chapter.prerequisites?.length ?? 0) > 0; else noPrereqs">
@@ -34,8 +42,9 @@
         <li *ngFor="let p of chapter.prerequisites">{{ describePrerequisite(p) }}</li>
       </ul>
       <small class="view-section-empty">
-        Pendant une partie, la quête se débloque dès que toutes ces conditions sont remplies.
+        Pendant une partie, {{ parentArc?.type === 'HUB' ? 'la quête' : 'ce chapitre' }} se débloque
-        Le toggle des faits se fait dans l'écran « Faits » de la Partie.
+        dès que toutes ces conditions sont remplies. Le toggle des faits se fait dans l'écran
+        « Faits » de la Partie.
       </small>
     </ng-container>
 

web/src/app/campaigns/chapter/chapter-view/chapter-view.component.scss

@@ -1,5 +1,21 @@
 // Styles partagés via styles/_view.scss
 
+// Badge "Conditionnel" : chapitre porteur de conditions de déblocage.
+.cond-badge {
+  display: inline-flex;
+  align-items: center;
+  gap: 0.25rem;
+  margin-left: 0.5rem;
+  padding: 0.1rem 0.5rem;
+  border-radius: 999px;
+  font-size: 0.72rem;
+  font-weight: 600;
+  vertical-align: middle;
+  color: #f0c674;
+  background: rgba(240, 198, 116, 0.12);
+  border: 1px solid rgba(240, 198, 116, 0.35);
+}
+
 .quest-status-bar {
   border: 1px solid rgba(66, 133, 244, 0.25);
   background: rgba(66, 133, 244, 0.04);

web/src/app/campaigns/chapter/chapter-view/chapter-view.component.ts

@@ -3,11 +3,12 @@ import { CommonModule } from '@angular/common';
 import { ActivatedRoute, Router, RouterModule } from '@angular/router';
 import { forkJoin, of } from 'rxjs';
 import { switchMap } from 'rxjs/operators';
-import { LucideAngularModule, Pencil, Network, Trash2 } from 'lucide-angular';
+import { LucideAngularModule, Pencil, Network, Trash2, Lock } from 'lucide-angular';
 import { resolveCampaignIcon } from '../../campaign-icons';
 import { CampaignService } from '../../../services/campaign.service';
 import { CharacterService } from '../../../services/character.service';
 import { NpcService } from '../../../services/npc.service';
+import { RandomTableService } from '../../../services/random-table.service';
 import { PageService } from '../../../services/page.service';
 import { LayoutService } from '../../../services/layout.service';
 import { PageTitleService } from '../../../services/page-title.service';
@@ -32,6 +33,7 @@ export class ChapterViewComponent implements OnInit, OnDestroy {
   readonly Pencil = Pencil;
   readonly Network = Network;
   readonly Trash2 = Trash2;
+  readonly Lock = Lock;
   readonly resolveCampaignIcon = resolveCampaignIcon;
 
   campaignId = '';
@@ -50,6 +52,7 @@ export class ChapterViewComponent implements OnInit, OnDestroy {
     private campaignService: CampaignService,
     private characterService: CharacterService,
     private npcService: NpcService,
+    private randomTableService: RandomTableService,
     private pageService: PageService,
     private layoutService: LayoutService,
     private pageTitleService: PageTitleService,
@@ -77,7 +80,7 @@ export class ChapterViewComponent implements OnInit, OnDestroy {
       campaign: this.campaignService.getCampaignById(this.campaignId),
       allCampaigns: this.campaignService.getAllCampaigns(),
       chapter: this.campaignService.getChapterById(this.chapterId),
-      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService)
+      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService, this.randomTableService)
     }).pipe(
       switchMap(data => {
         const lid = data.campaign.loreId ?? null;

web/src/app/campaigns/notebook/notebook-action-card/notebook-action-card.component.html

@@ -0,0 +1,41 @@
+<div class="nac" [class.created]="status === 'created'">
+  <div class="nac-head">
+    <lucide-icon [img]="typeIcon" [size]="15"></lucide-icon>
+    <span class="nac-type">{{ typeLabel }}</span>
+    <span class="nac-name">{{ action.name }}</span>
+  </div>
+
+  <p class="nac-desc" *ngIf="action.description">{{ action.description }}</p>
+
+  <!-- Cibles -->
+  <div class="nac-targets" *ngIf="status !== 'created' && needsArc">
+    <label>
+      Arc
+      <select [(ngModel)]="selectedArcId" (ngModelChange)="syncChapter()">
+        <option *ngFor="let a of arcs" [value]="a.id">{{ a.name }}</option>
+      </select>
+    </label>
+    <label *ngIf="needsChapter">
+      Chapitre
+      <select [(ngModel)]="selectedChapterId">
+        <option *ngFor="let c of targetChapters" [value]="c.id">{{ c.name }}</option>
+      </select>
+    </label>
+  </div>
+
+  <p class="nac-warn" *ngIf="needsArc && arcs.length === 0">
+    Crée d'abord un arc {{ needsChapter ? 'et un chapitre ' : '' }}dans la campagne pour pouvoir y rattacher cet élément.
+  </p>
+
+  <div class="nac-foot">
+    <button class="nac-create" *ngIf="status !== 'created'" (click)="create()" [disabled]="!canCreate">
+      <lucide-icon [img]="Plus" [size]="13"></lucide-icon>
+      {{ status === 'creating' ? 'Création…' : 'Créer dans la campagne' }}
+    </button>
+    <button class="nac-open" *ngIf="status === 'created'" (click)="openCreated()">
+      <lucide-icon [img]="Check" [size]="13"></lucide-icon> Créé — ouvrir
+      <lucide-icon [img]="ExternalLink" [size]="12"></lucide-icon>
+    </button>
+    <span class="nac-error" *ngIf="status === 'error'">{{ errorMsg }}</span>
+  </div>
+</div>

web/src/app/campaigns/notebook/notebook-action-card/notebook-action-card.component.scss

@@ -0,0 +1,52 @@
+.nac {
+  margin-top: 0.5rem;
+  padding: 0.6rem 0.75rem;
+  border-radius: 9px;
+  border: 1px solid rgba(168, 130, 255, 0.3);
+  background: rgba(168, 130, 255, 0.08);
+
+  &.created { border-color: rgba(107, 208, 138, 0.4); background: rgba(107, 208, 138, 0.08); }
+}
+
+.nac-head {
+  display: flex; align-items: center; gap: 0.4rem;
+  color: #c4a8ff;
+  .nac-type {
+    font-size: 0.68rem; text-transform: uppercase; letter-spacing: 0.05em;
+    padding: 0.05rem 0.4rem; border-radius: 4px; background: rgba(168,130,255,0.18);
+  }
+  .nac-name { font-weight: 600; color: inherit; }
+}
+
+.nac-desc {
+  margin: 0.4rem 0 0; font-size: 0.85rem; color: var(--color-text-muted, #cfd3da);
+  white-space: pre-wrap;
+}
+
+.nac-targets {
+  display: flex; gap: 0.6rem; flex-wrap: wrap; margin-top: 0.5rem;
+  label { display: flex; flex-direction: column; gap: 0.2rem; font-size: 0.72rem; color: var(--color-text-muted, #9aa0aa); }
+  select {
+    padding: 0.3rem 0.45rem; border-radius: 6px; font: inherit;
+    border: 1px solid rgba(255,255,255,0.14); background: rgba(255,255,255,0.04); color: inherit;
+  }
+}
+
+.nac-warn { margin: 0.4rem 0 0; font-size: 0.78rem; color: #e0a458; }
+
+.nac-foot { display: flex; align-items: center; gap: 0.6rem; margin-top: 0.55rem; }
+
+.nac-create, .nac-open {
+  display: inline-flex; align-items: center; gap: 0.35rem;
+  padding: 0.35rem 0.7rem; border-radius: 7px; cursor: pointer; font-size: 0.82rem; font-weight: 600;
+  border: none;
+}
+.nac-create {
+  background: #8a6dff; color: #fff;
+  &:hover:not(:disabled) { background: #7a5cf0; }
+  &:disabled { opacity: 0.5; cursor: default; }
+}
+.nac-open { background: rgba(107,208,138,0.2); color: #6bd08a; }
+.nac-open:hover { background: rgba(107,208,138,0.3); }
+
+.nac-error { color: #e88; font-size: 0.8rem; }

web/src/app/campaigns/notebook/notebook-action-card/notebook-action-card.component.ts

@@ -0,0 +1,185 @@
+import { Component, EventEmitter, Input, OnInit, Output } from '@angular/core';
+import { CommonModule } from '@angular/common';
+import { FormsModule } from '@angular/forms';
+import { Router } from '@angular/router';
+import { LucideAngularModule, Plus, Check, Drama, Clapperboard, BookText, GitBranch, Dices, ExternalLink } from 'lucide-angular';
+import { CampaignService } from '../../../services/campaign.service';
+import { NpcService } from '../../../services/npc.service';
+import { RandomTableService } from '../../../services/random-table.service';
+import { Arc, Chapter } from '../../../services/campaign.model';
+import { NotebookAction } from '../../../services/notebook-action.model';
+
+/**
+ * Carte « Créer dans la campagne » issue d'une proposition de l'IA. Gère la cible
+ * (chapitre pour une scène, arc pour un chapitre) et appelle les services existants.
+ */
+@Component({
+  selector: 'app-notebook-action-card',
+  standalone: true,
+  imports: [CommonModule, FormsModule, LucideAngularModule],
+  templateUrl: './notebook-action-card.component.html',
+  styleUrls: ['./notebook-action-card.component.scss']
+})
+export class NotebookActionCardComponent implements OnInit {
+  readonly Plus = Plus;
+  readonly Check = Check;
+  readonly ExternalLink = ExternalLink;
+
+  @Input() action!: NotebookAction;
+  @Input() campaignId!: string;
+  @Input() arcs: Arc[] = [];
+  @Input() chaptersByArc: Record<string, Chapter[]> = {};
+  /** Émis après une création réussie → l'atelier rafraîchit la sidebar. */
+  @Output() created = new EventEmitter<void>();
+
+  selectedArcId = '';
+  selectedChapterId = '';
+
+  status: 'idle' | 'creating' | 'created' | 'error' = 'idle';
+  errorMsg = '';
+  createdRoute: string[] | null = null;
+
+  constructor(
+    private campaignService: CampaignService,
+    private npcService: NpcService,
+    private tableService: RandomTableService,
+    private router: Router
+  ) {}
+
+  ngOnInit(): void {
+    if (this.arcs.length) this.selectedArcId = this.arcs[0].id!;
+    this.syncChapter();
+  }
+
+  get typeLabel(): string {
+    switch (this.action.type) {
+      case 'npc': return 'PNJ';
+      case 'scene': return 'Scène';
+      case 'chapter': return 'Chapitre';
+      case 'arc': return 'Arc';
+      case 'table': return 'Table aléatoire';
+      default: return this.action.type;
+    }
+  }
+
+  get typeIcon() {
+    switch (this.action.type) {
+      case 'npc': return Drama;
+      case 'scene': return Clapperboard;
+      case 'chapter': return BookText;
+      case 'arc': return GitBranch;
+      case 'table': return Dices;
+      default: return Plus;
+    }
+  }
+
+  get needsArc(): boolean { return this.action.type === 'chapter' || this.action.type === 'scene'; }
+  get needsChapter(): boolean { return this.action.type === 'scene'; }
+
+  get targetChapters(): Chapter[] { return this.chaptersByArc[this.selectedArcId] ?? []; }
+
+  syncChapter(): void {
+    const chs = this.targetChapters;
+    this.selectedChapterId = chs.length ? chs[0].id! : '';
+  }
+
+  get canCreate(): boolean {
+    if (this.status === 'creating' || this.status === 'created') return false;
+    if (this.needsArc && !this.selectedArcId) return false;
+    if (this.needsChapter && !this.selectedChapterId) return false;
+    return true;
+  }
+
+  create(): void {
+    if (!this.canCreate) return;
+    this.status = 'creating';
+    this.errorMsg = '';
+    switch (this.action.type) {
+      case 'npc':     return this.createNpc();
+      case 'arc':     return this.createArc();
+      case 'chapter': return this.createChapter();
+      case 'scene':   return this.createScene();
+      case 'table':   return this.createTable();
+    }
+  }
+
+  private createNpc(): void {
+    this.npcService.create({
+      name: this.action.name,
+      campaignId: this.campaignId,
+      values: this.action.description ? { Description: this.action.description } : {}
+    }).subscribe({
+      next: (n) => this.done(['/campaigns', this.campaignId, 'npcs', n.id!]),
+      error: (e) => this.fail(e)
+    });
+  }
+
+  private createArc(): void {
+    this.campaignService.createArc({
+      name: this.action.name,
+      description: this.action.description,
+      campaignId: this.campaignId,
+      order: this.arcs.length,
+      type: this.action.arcType === 'HUB' ? 'HUB' : 'LINEAR'
+    }).subscribe({
+      next: (a) => this.done(['/campaigns', this.campaignId, 'arcs', a.id!]),
+      error: (e) => this.fail(e)
+    });
+  }
+
+  private createChapter(): void {
+    const order = (this.chaptersByArc[this.selectedArcId] ?? []).length;
+    this.campaignService.createChapter({
+      name: this.action.name,
+      description: this.action.description,
+      arcId: this.selectedArcId,
+      order
+    }).subscribe({
+      next: (c) => this.done(['/campaigns', this.campaignId, 'arcs', this.selectedArcId, 'chapters', c.id!]),
+      error: (e) => this.fail(e)
+    });
+  }
+
+  private createScene(): void {
+    this.campaignService.createScene({
+      name: this.action.name,
+      description: this.action.description,
+      playerNarration: this.action.content,
+      chapterId: this.selectedChapterId,
+      order: 0
+    }).subscribe({
+      next: (s) => this.done(
+        ['/campaigns', this.campaignId, 'arcs', this.selectedArcId, 'chapters', this.selectedChapterId, 'scenes', s.id!]),
+      error: (e) => this.fail(e)
+    });
+  }
+
+  private createTable(): void {
+    this.tableService.create({
+      name: this.action.name,
+      diceFormula: this.action.diceFormula || '1d20',
+      campaignId: this.campaignId,
+      entries: (this.action.entries ?? []).map(e => ({
+        minRoll: e.minRoll, maxRoll: e.maxRoll, label: e.label, detail: e.detail
+      }))
+    }).subscribe({
+      next: (t) => this.done(['/campaigns', this.campaignId, 'random-tables', t.id!]),
+      error: (e) => this.fail(e)
+    });
+  }
+
+  private done(route: string[]): void {
+    this.status = 'created';
+    this.createdRoute = route;
+    this.created.emit();
+  }
+
+  private fail(err: unknown): void {
+    this.status = 'error';
+    this.errorMsg = (err as { error?: { message?: string } })?.error?.message || 'Échec de la création.';
+  }
+
+  openCreated(): void {
+    if (this.createdRoute) this.router.navigate(this.createdRoute);
+  }
+}

web/src/app/campaigns/notebook/notebook-detail/notebook-detail.component.html

@@ -0,0 +1,98 @@
+<div class="nbd-page" *ngIf="detail">
+  <div class="nbd-toolbar">
+    <button class="btn-back" (click)="back()">
+      <lucide-icon [img]="ArrowLeft" [size]="14"></lucide-icon> Ateliers
+    </button>
+    <input class="nbd-title" [(ngModel)]="detail.name" (blur)="rename()" (keyup.enter)="rename()">
+  </div>
+
+  <div class="nbd-grid">
+    <!-- Sources -->
+    <aside class="nbd-sources">
+      <div class="nbd-sources-head">
+        <h3><lucide-icon [img]="FileText" [size]="15"></lucide-icon> Sources</h3>
+        <label class="btn-upload" [class.disabled]="uploading">
+          <lucide-icon [img]="Upload" [size]="13"></lucide-icon>
+          {{ uploading ? 'Indexation…' : 'Ajouter un PDF' }}
+          <input type="file" accept="application/pdf" hidden (change)="onFile($event)" [disabled]="uploading">
+        </label>
+      </div>
+
+      <p class="nbd-upload-error" *ngIf="uploadError">{{ uploadError }}</p>
+
+      <div class="nbd-source" *ngFor="let s of sources">
+        <lucide-icon
+          [img]="s.status === 'READY' ? CheckCircle2 : (s.status === 'FAILED' ? AlertCircle : Loader)"
+          [size]="14"
+          [class.ok]="s.status === 'READY'" [class.fail]="s.status === 'FAILED'" [class.busy]="s.status === 'INDEXING'">
+        </lucide-icon>
+        <div class="nbd-source-info">
+          <div class="nbd-source-name">{{ s.filename }}</div>
+          <div class="nbd-source-meta">
+            <span *ngIf="s.status === 'READY'">{{ s.pageCount }} p. · {{ s.chunkCount }} extraits</span>
+            <span *ngIf="s.status === 'INDEXING'">indexation en cours…</span>
+            <span *ngIf="s.status === 'FAILED'">échec de l'indexation</span>
+          </div>
+        </div>
+        <button class="nbd-source-del" (click)="removeSource(s)" title="Supprimer">
+          <lucide-icon [img]="Trash2" [size]="13"></lucide-icon>
+        </button>
+      </div>
+
+      <p class="nbd-empty" *ngIf="sources.length === 0 && !uploading">
+        Ajoute un PDF source pour commencer à discuter avec.
+      </p>
+    </aside>
+
+    <!-- Chat -->
+    <section class="nbd-chat">
+      <div class="nbd-messages">
+        <p class="nbd-empty" *ngIf="messages.length === 0">
+          Pose une question sur ta source, ou demande une adaptation pour ta campagne.
+          <span *ngIf="!hasReadySource()"><br>(Ajoute d'abord une source indexée pour des réponses ancrées.)</span>
+        </p>
+        <div class="nbd-msg" *ngFor="let m of messages" [class.user]="m.role === 'user'" [class.assistant]="m.role === 'assistant'">
+          <div class="nbd-msg-role">
+            <lucide-icon *ngIf="m.role === 'assistant'" [img]="Sparkles" [size]="12"></lucide-icon>
+            {{ m.role === 'user' ? 'Vous' : 'IA' }}
+          </div>
+
+          <ng-container *ngIf="m.role === 'assistant'; else userContent">
+            <ng-container *ngIf="parsedOf(m) as p">
+              <div class="nbd-deep-progress" *ngIf="sending && deepProgress && !p.text">
+                <lucide-icon [img]="Layers" [size]="13"></lucide-icon>
+                Analyse approfondie du document… {{ deepProgress.current }}/{{ deepProgress.total }}
+              </div>
+              <div class="nbd-msg-content">{{ p.text }}<span class="cursor" *ngIf="sending && !p.text && !p.actions.length && !deepProgress">▌</span></div>
+              <app-notebook-action-card
+                *ngFor="let a of p.actions; trackBy: trackAction"
+                [action]="a"
+                [campaignId]="campaignId"
+                [arcs]="arcs"
+                [chaptersByArc]="chaptersByArc"
+                (created)="onActionCreated()">
+              </app-notebook-action-card>
+            </ng-container>
+          </ng-container>
+          <ng-template #userContent>
+            <div class="nbd-msg-content">{{ m.content }}</div>
+          </ng-template>
+        </div>
+      </div>
+
+      <div class="nbd-input">
+        <textarea [(ngModel)]="draft" rows="2"
+                  placeholder="Demande une adaptation, un résumé, un PNJ inspiré de la source…"
+                  (keydown.enter)="$event.preventDefault(); send()"></textarea>
+        <button class="btn-deep" (click)="send(true)" [disabled]="sending || !draft.trim()"
+                title="Analyse approfondie : lit TOUT le document (plus lent, exhaustif). Idéal pour « liste tous les… »">
+          <lucide-icon [img]="Layers" [size]="15"></lucide-icon>
+        </button>
+        <button class="btn-send" (click)="send()" [disabled]="sending || !draft.trim()"
+                title="Réponse rapide (recherche ciblée dans le document)">
+          <lucide-icon [img]="Send" [size]="15"></lucide-icon>
+        </button>
+      </div>
+    </section>
+  </div>
+</div>

web/src/app/campaigns/notebook/notebook-detail/notebook-detail.component.scss

@@ -0,0 +1,106 @@
+.nbd-page { max-width: 1100px; margin: 0 auto; padding: 1rem 1.5rem 2rem; height: calc(100vh - 60px); display: flex; flex-direction: column; }
+
+.nbd-toolbar {
+  display: flex; align-items: center; gap: 0.75rem; margin-bottom: 1rem;
+  .btn-back {
+    display: inline-flex; align-items: center; gap: 0.35rem; flex-shrink: 0;
+    padding: 0.4rem 0.7rem; border-radius: 6px;
+    border: 1px solid rgba(255,255,255,0.12); background: rgba(255,255,255,0.04);
+    color: inherit; cursor: pointer; font-size: 0.85rem;
+    &:hover { background: rgba(255,255,255,0.09); }
+  }
+  .nbd-title {
+    flex: 1; padding: 0.4rem 0.6rem; border-radius: 6px; font-size: 1.15rem; font-weight: 600;
+    border: 1px solid transparent; background: transparent; color: inherit;
+    &:hover, &:focus { border-color: rgba(255,255,255,0.14); background: rgba(255,255,255,0.04); outline: none; }
+  }
+}
+
+.nbd-grid {
+  flex: 1; min-height: 0;
+  display: grid; grid-template-columns: 280px 1fr; gap: 1rem;
+}
+
+/* Sources */
+.nbd-sources {
+  display: flex; flex-direction: column; gap: 0.5rem; overflow-y: auto;
+  border: 1px solid rgba(255,255,255,0.08); border-radius: 10px; padding: 0.75rem;
+}
+.nbd-sources-head {
+  display: flex; align-items: center; justify-content: space-between; gap: 0.5rem;
+  h3 { display: flex; align-items: center; gap: 0.4rem; margin: 0; font-size: 0.95rem; }
+}
+.btn-upload {
+  display: inline-flex; align-items: center; gap: 0.3rem; cursor: pointer;
+  padding: 0.3rem 0.55rem; border-radius: 6px; font-size: 0.78rem;
+  border: 1px solid rgba(102,126,234,0.4); color: #8ea2ff; background: rgba(102,126,234,0.1);
+  &:hover { background: rgba(102,126,234,0.2); }
+  &.disabled { opacity: 0.6; cursor: default; }
+}
+.nbd-upload-error { color: #e88; font-size: 0.8rem; margin: 0.2rem 0; }
+
+.nbd-source {
+  display: flex; align-items: flex-start; gap: 0.5rem; padding: 0.5rem;
+  border-radius: 7px; background: rgba(255,255,255,0.03);
+  lucide-icon.ok { color: #6bd08a; }
+  lucide-icon.fail { color: #e88; }
+  lucide-icon.busy { color: #e0c074; }
+  .nbd-source-info { flex: 1; min-width: 0; }
+  .nbd-source-name { font-size: 0.85rem; font-weight: 500; word-break: break-word; }
+  .nbd-source-meta { font-size: 0.72rem; color: var(--color-text-muted, #9aa0aa); }
+  .nbd-source-del {
+    border: none; background: none; color: #e88; cursor: pointer; padding: 0.15rem;
+    border-radius: 4px; &:hover { background: rgba(224,90,90,0.15); }
+  }
+}
+
+/* Chat */
+.nbd-chat {
+  display: flex; flex-direction: column; min-height: 0;
+  border: 1px solid rgba(255,255,255,0.08); border-radius: 10px;
+}
+.nbd-messages { flex: 1; overflow-y: auto; padding: 1rem; display: flex; flex-direction: column; gap: 0.85rem; }
+.nbd-empty { color: var(--color-text-muted, #9aa0aa); font-style: italic; }
+
+.nbd-msg {
+  max-width: 88%;
+  .nbd-msg-role {
+    display: flex; align-items: center; gap: 0.3rem;
+    font-size: 0.7rem; text-transform: uppercase; letter-spacing: 0.04em;
+    color: var(--color-text-muted, #9aa0aa); margin-bottom: 0.2rem;
+  }
+  .nbd-msg-content { white-space: pre-wrap; line-height: 1.5; }
+  &.user { align-self: flex-end; text-align: right;
+    .nbd-msg-content { background: rgba(102,126,234,0.16); padding: 0.5rem 0.75rem; border-radius: 10px; display: inline-block; text-align: left; }
+  }
+  &.assistant { align-self: flex-start;
+    .nbd-msg-content { background: rgba(255,255,255,0.04); padding: 0.5rem 0.75rem; border-radius: 10px; }
+  }
+}
+.cursor { opacity: 0.6; }
+
+.nbd-input {
+  display: flex; gap: 0.5rem; padding: 0.6rem; border-top: 1px solid rgba(255,255,255,0.08);
+  textarea {
+    flex: 1; resize: none; padding: 0.55rem 0.7rem; border-radius: 8px;
+    border: 1px solid rgba(255,255,255,0.14); background: rgba(255,255,255,0.04); color: inherit; font: inherit;
+  }
+  .btn-send {
+    display: inline-flex; align-items: center; justify-content: center; width: 44px;
+    border: none; border-radius: 8px; background: #667eea; color: #fff; cursor: pointer;
+    &:hover:not(:disabled) { background: #5568d3; }
+    &:disabled { opacity: 0.5; cursor: default; }
+  }
+  .btn-deep {
+    display: inline-flex; align-items: center; justify-content: center; width: 44px;
+    border: 1px solid rgba(168,130,255,0.4); border-radius: 8px;
+    background: rgba(168,130,255,0.12); color: #c4a8ff; cursor: pointer;
+    &:hover:not(:disabled) { background: rgba(168,130,255,0.22); }
+    &:disabled { opacity: 0.5; cursor: default; }
+  }
+}
+
+.nbd-deep-progress {
+  display: inline-flex; align-items: center; gap: 0.35rem;
+  font-size: 0.82rem; color: #c4a8ff; font-style: italic;
+}

web/src/app/campaigns/notebook/notebook-detail/notebook-detail.component.ts

@@ -0,0 +1,187 @@
+import { Component, OnInit } from '@angular/core';
+import { CommonModule } from '@angular/common';
+import { FormsModule } from '@angular/forms';
+import { ActivatedRoute, Router } from '@angular/router';
+import { LucideAngularModule, ArrowLeft, Upload, Trash2, Send, FileText, Loader, CheckCircle2, AlertCircle, Sparkles, Layers } from 'lucide-angular';
+import { NotebookService } from '../../../services/notebook.service';
+import { CampaignSidebarService } from '../../../services/campaign-sidebar.service';
+import { CampaignService } from '../../../services/campaign.service';
+import { CharacterService } from '../../../services/character.service';
+import { NpcService } from '../../../services/npc.service';
+import { NotebookDetail, NotebookSource, NotebookMessage } from '../../../services/notebook.model';
+import { parseNotebookActions, NotebookAction } from '../../../services/notebook-action.model';
+import { Arc, Chapter } from '../../../services/campaign.model';
+import { loadCampaignTreeData } from '../../campaign-tree.helper';
+import { NotebookActionCardComponent } from '../notebook-action-card/notebook-action-card.component';
+
+/**
+ * Atelier (workspace) : sources indexées (gauche) + chat ancré RAG (droite).
+ * Route : /campaigns/:campaignId/notebooks/:notebookId
+ */
+@Component({
+  selector: 'app-notebook-detail',
+  standalone: true,
+  imports: [CommonModule, FormsModule, LucideAngularModule, NotebookActionCardComponent],
+  templateUrl: './notebook-detail.component.html',
+  styleUrls: ['./notebook-detail.component.scss']
+})
+export class NotebookDetailComponent implements OnInit {
+  readonly ArrowLeft = ArrowLeft;
+  readonly Upload = Upload;
+  readonly Trash2 = Trash2;
+  readonly Send = Send;
+  readonly FileText = FileText;
+  readonly Loader = Loader;
+  readonly CheckCircle2 = CheckCircle2;
+  readonly AlertCircle = AlertCircle;
+  readonly Sparkles = Sparkles;
+  readonly Layers = Layers;
+
+  campaignId = '';
+  notebookId = '';
+  detail: NotebookDetail | null = null;
+  sources: NotebookSource[] = [];
+  messages: NotebookMessage[] = [];
+
+  uploading = false;
+  uploadError = '';
+  sending = false;
+  draft = '';
+  /** Avancement de l'analyse approfondie (lecture du doc par lots). */
+  deepProgress: { current: number; total: number } | null = null;
+
+  // Arbre de la campagne — sert aux cartes d'action (cibles arc/chapitre).
+  arcs: Arc[] = [];
+  chaptersByArc: Record<string, Chapter[]> = {};
+
+  constructor(
+    private route: ActivatedRoute,
+    private router: Router,
+    private service: NotebookService,
+    private campaignSidebar: CampaignSidebarService,
+    private campaignService: CampaignService,
+    private characterService: CharacterService,
+    private npcService: NpcService
+  ) {}
+
+  ngOnInit(): void {
+    this.campaignId = this.route.snapshot.paramMap.get('campaignId') ?? '';
+    this.notebookId = this.route.snapshot.paramMap.get('notebookId') ?? '';
+    if (this.campaignId) {
+      this.campaignSidebar.show(this.campaignId);
+      this.loadTree();
+    }
+    this.load();
+  }
+
+  private loadTree(): void {
+    loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService)
+      .subscribe({
+        next: (data) => { this.arcs = data.arcs; this.chaptersByArc = data.chaptersByArc; },
+        error: () => { /* cibles indisponibles : les cartes le signaleront */ }
+      });
+  }
+
+  /** Après création depuis une carte : rafraîchit la sidebar (l'élément apparaît)
+   *  et l'arbre des cibles (pour les cartes suivantes). */
+  onActionCreated(): void {
+    if (this.campaignId) {
+      this.campaignSidebar.show(this.campaignId);
+      this.loadTree();
+    }
+  }
+
+  /** Sépare le texte affiché des blocs d'action — MÉMORISÉ par message : renvoie
+   *  une référence STABLE tant que le contenu n'a pas changé. Indispensable :
+   *  appeler le parseur directement dans le template recréait le DOM à chaque
+   *  détection de changement (au mousedown) → les clics sur les cartes étaient
+   *  perdus. */
+  parsedOf(m: NotebookMessage): { text: string; actions: NotebookAction[] } {
+    const cache = m as unknown as {
+      _parsedFor?: string;
+      _parsed?: { text: string; actions: NotebookAction[] };
+    };
+    if (cache._parsedFor !== m.content || !cache._parsed) {
+      cache._parsed = parseNotebookActions(m.content);
+      cache._parsedFor = m.content;
+    }
+    return cache._parsed;
+  }
+
+  /** trackBy stable pour les cartes d'action (évite toute recréation parasite). */
+  trackAction(index: number): number { return index; }
+
+  load(): void {
+    this.service.get(this.notebookId).subscribe({
+      next: (d) => {
+        this.detail = d;
+        this.sources = d.sources ?? [];
+        this.messages = d.messages ?? [];
+      },
+      error: () => this.back()
+    });
+  }
+
+  reloadSources(): void {
+    this.service.get(this.notebookId).subscribe({ next: (d) => this.sources = d.sources ?? [] });
+  }
+
+  // --- Sources ---
+
+  onFile(event: Event): void {
+    const input = event.target as HTMLInputElement;
+    const file = input.files?.[0];
+    input.value = '';
+    if (!file) return;
+    this.uploading = true;
+    this.uploadError = '';
+    this.service.addSource(this.notebookId, file).subscribe({
+      next: () => { this.uploading = false; this.reloadSources(); },
+      error: (err) => {
+        this.uploading = false;
+        this.uploadError = err?.error?.message || 'Échec de l\'indexation. Réessaie ou vérifie le modèle d\'embedding.';
+        this.reloadSources();
+      }
+    });
+  }
+
+  removeSource(s: NotebookSource): void {
+    this.service.deleteSource(s.id).subscribe(() => this.reloadSources());
+  }
+
+  // --- Chat ---
+
+  send(deep = false): void {
+    const text = this.draft.trim();
+    if (!text || this.sending) return;
+    this.draft = '';
+    this.deepProgress = null;
+    this.messages.push({ role: 'user', content: text });
+    const assistant: NotebookMessage = { role: 'assistant', content: '' };
+    this.messages.push(assistant);
+    this.sending = true;
+
+    this.service.streamChat(this.notebookId, text, deep).subscribe({
+      next: (ev) => {
+        if (ev.type === 'token') { this.deepProgress = null; assistant.content += ev.value; }
+        else if (ev.type === 'progress') this.deepProgress = { current: ev.current, total: ev.total };
+        else if (ev.type === 'error') assistant.content += (assistant.content ? '\n\n' : '') + `⚠️ ${ev.message}`;
+      },
+      complete: () => { this.sending = false; this.deepProgress = null; },
+      error: () => { this.sending = false; this.deepProgress = null; }
+    });
+  }
+
+  rename(): void {
+    if (!this.detail || !this.detail.name.trim()) return;
+    this.service.rename(this.notebookId, this.detail.name.trim()).subscribe();
+  }
+
+  back(): void {
+    this.router.navigate(['/campaigns', this.campaignId, 'notebooks']);
+  }
+
+  hasReadySource(): boolean {
+    return this.sources.some(s => s.status === 'READY');
+  }
+}

web/src/app/campaigns/notebook/notebook-list/notebook-list.component.html

@@ -0,0 +1,35 @@
+<div class="nbl-page">
+  <div class="nbl-toolbar">
+    <button class="btn-back" (click)="back()">
+      <lucide-icon [img]="ArrowLeft" [size]="14"></lucide-icon> Retour
+    </button>
+  </div>
+
+  <header class="nbl-header">
+    <h1><lucide-icon [img]="BookOpen" [size]="22"></lucide-icon> Ateliers d'adaptation</h1>
+    <p class="nbl-hint">
+      Discute avec l'IA à partir d'un PDF source (recherche dans le document) pour adapter
+      son contenu à ta campagne. La source et la conversation sont conservées.
+    </p>
+  </header>
+
+  <div class="nbl-create">
+    <input type="text" [(ngModel)]="newName" placeholder="Nom de l'atelier (ex: Adaptation du module X)"
+           (keyup.enter)="create()">
+    <button class="btn-create" (click)="create()" [disabled]="creating">
+      <lucide-icon [img]="Plus" [size]="14"></lucide-icon>
+      {{ creating ? 'Création…' : 'Nouvel atelier' }}
+    </button>
+  </div>
+
+  <div class="nbl-list">
+    <p class="empty" *ngIf="notebooks.length === 0">Aucun atelier pour l'instant.</p>
+    <button class="nbl-item" *ngFor="let nb of notebooks" (click)="open(nb)">
+      <lucide-icon [img]="BookOpen" [size]="16"></lucide-icon>
+      <span class="nbl-item-name">{{ nb.name }}</span>
+      <span class="nbl-del" (click)="remove(nb, $event)" title="Supprimer">
+        <lucide-icon [img]="Trash2" [size]="14"></lucide-icon>
+      </span>
+    </button>
+  </div>
+</div>

web/src/app/campaigns/notebook/notebook-list/notebook-list.component.scss

@@ -0,0 +1,49 @@
+.nbl-page { max-width: 760px; margin: 0 auto; padding: 1rem 1.5rem 3rem; }
+
+.nbl-toolbar { margin-bottom: 1rem; }
+.btn-back {
+  display: inline-flex; align-items: center; gap: 0.35rem;
+  padding: 0.4rem 0.7rem; border-radius: 6px;
+  border: 1px solid rgba(255,255,255,0.12); background: rgba(255,255,255,0.04);
+  color: inherit; cursor: pointer; font-size: 0.85rem;
+  &:hover { background: rgba(255,255,255,0.09); }
+}
+
+.nbl-header {
+  margin-bottom: 1.25rem;
+  h1 { display: flex; align-items: center; gap: 0.5rem; margin: 0 0 0.35rem; font-size: 1.5rem; }
+  .nbl-hint { margin: 0; color: var(--color-text-muted, #9aa0aa); font-size: 0.9rem; }
+}
+
+.nbl-create {
+  display: flex; gap: 0.5rem; margin-bottom: 1.5rem;
+  input {
+    flex: 1; padding: 0.55rem 0.75rem; border-radius: 7px;
+    border: 1px solid rgba(255,255,255,0.14); background: rgba(255,255,255,0.04);
+    color: inherit; font: inherit;
+  }
+  .btn-create {
+    display: inline-flex; align-items: center; gap: 0.4rem;
+    padding: 0.55rem 1rem; border: none; border-radius: 7px;
+    background: #667eea; color: #fff; font-weight: 600; cursor: pointer;
+    &:hover:not(:disabled) { background: #5568d3; }
+    &:disabled { opacity: 0.55; cursor: default; }
+  }
+}
+
+.nbl-list { display: flex; flex-direction: column; gap: 0.4rem; }
+.empty { color: var(--color-text-muted, #9aa0aa); font-style: italic; }
+
+.nbl-item {
+  display: flex; align-items: center; gap: 0.55rem;
+  padding: 0.7rem 0.85rem; border-radius: 8px;
+  border: 1px solid rgba(255,255,255,0.1); background: rgba(255,255,255,0.03);
+  color: inherit; cursor: pointer; text-align: left;
+  &:hover { background: rgba(255,255,255,0.07); }
+
+  .nbl-item-name { flex: 1; font-weight: 500; }
+  .nbl-del {
+    display: inline-flex; padding: 0.25rem; border-radius: 5px; color: #e88;
+    &:hover { background: rgba(224,90,90,0.15); }
+  }
+}

web/src/app/campaigns/notebook/notebook-list/notebook-list.component.ts

@@ -0,0 +1,85 @@
+import { Component, OnInit } from '@angular/core';
+import { CommonModule } from '@angular/common';
+import { FormsModule } from '@angular/forms';
+import { ActivatedRoute, Router } from '@angular/router';
+import { LucideAngularModule, ArrowLeft, Plus, Trash2, BookOpen } from 'lucide-angular';
+import { NotebookService } from '../../../services/notebook.service';
+import { CampaignSidebarService } from '../../../services/campaign-sidebar.service';
+import { Notebook } from '../../../services/notebook.model';
+import { ConfirmDialogService } from '../../../shared/confirm-dialog/confirm-dialog.service';
+
+/**
+ * Liste des ateliers (notebooks) d'une campagne + création.
+ * Route : /campaigns/:campaignId/notebooks
+ */
+@Component({
+  selector: 'app-notebook-list',
+  standalone: true,
+  imports: [CommonModule, FormsModule, LucideAngularModule],
+  templateUrl: './notebook-list.component.html',
+  styleUrls: ['./notebook-list.component.scss']
+})
+export class NotebookListComponent implements OnInit {
+  readonly ArrowLeft = ArrowLeft;
+  readonly Plus = Plus;
+  readonly Trash2 = Trash2;
+  readonly BookOpen = BookOpen;
+
+  campaignId = '';
+  notebooks: Notebook[] = [];
+  newName = '';
+  creating = false;
+
+  constructor(
+    private route: ActivatedRoute,
+    private router: Router,
+    private service: NotebookService,
+    private campaignSidebar: CampaignSidebarService,
+    private confirmDialog: ConfirmDialogService
+  ) {}
+
+  ngOnInit(): void {
+    this.campaignId = this.route.snapshot.paramMap.get('campaignId') ?? '';
+    if (this.campaignId) {
+      this.campaignSidebar.show(this.campaignId);
+      this.load();
+    }
+  }
+
+  load(): void {
+    this.service.listByCampaign(this.campaignId).subscribe({
+      next: (list) => this.notebooks = list,
+      error: () => this.notebooks = []
+    });
+  }
+
+  create(): void {
+    if (this.creating) return;
+    this.creating = true;
+    this.service.create(this.campaignId, this.newName.trim() || 'Nouvel atelier').subscribe({
+      next: (nb) => this.router.navigate(['/campaigns', this.campaignId, 'notebooks', nb.id]),
+      error: () => this.creating = false
+    });
+  }
+
+  open(nb: Notebook): void {
+    this.router.navigate(['/campaigns', this.campaignId, 'notebooks', nb.id]);
+  }
+
+  remove(nb: Notebook, ev: Event): void {
+    ev.stopPropagation();
+    this.confirmDialog.confirm({
+      title: 'Supprimer l\'atelier',
+      message: `Supprimer « ${nb.name} » et ses sources indexées ?`,
+      confirmLabel: 'Supprimer',
+      variant: 'danger'
+    }).then(ok => {
+      if (!ok) return;
+      this.service.delete(nb.id).subscribe(() => this.load());
+    });
+  }
+
+  back(): void {
+    this.router.navigate(['/campaigns', this.campaignId]);
+  }
+}

web/src/app/campaigns/playthrough/playthrough-detail/playthrough-detail.component.ts

@@ -7,6 +7,7 @@ import { LucideAngularModule, ArrowLeft, Play, Flag, Users, Trash2, Pencil, Plus
 import { CampaignService } from '../../../services/campaign.service';
 import { CharacterService } from '../../../services/character.service';
 import { NpcService } from '../../../services/npc.service';
+import { RandomTableService } from '../../../services/random-table.service';
 import { PlaythroughService } from '../../../services/playthrough.service';
 import { SessionService } from '../../../services/session.service';
 import { LayoutService } from '../../../services/layout.service';
@@ -55,6 +56,7 @@ export class PlaythroughDetailComponent implements OnInit, OnDestroy {
     private campaignService: CampaignService,
     private characterService: CharacterService,
     private npcService: NpcService,
+    private randomTableService: RandomTableService,
     private playthroughService: PlaythroughService,
     private sessionService: SessionService,
     private layoutService: LayoutService,
@@ -78,7 +80,7 @@ export class PlaythroughDetailComponent implements OnInit, OnDestroy {
     forkJoin({
       campaign: this.campaignService.getCampaignById(this.campaignId),
       allCampaigns: this.campaignService.getAllCampaigns(),
-      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService),
+      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService, this.randomTableService),
       playthrough: this.playthroughService.getById(this.playthroughId),
       sessions: this.sessionService.getSessions(this.playthroughId).pipe(catchError(() => of([] as Session[]))),
       characters: this.characterService.getByPlaythrough(this.playthroughId).pipe(catchError(() => of([] as Character[]))),

web/src/app/campaigns/playthrough/playthrough-flags-page/playthrough-flags-page.component.ts

@@ -6,6 +6,7 @@ import { LucideAngularModule, ArrowLeft } from 'lucide-angular';
 import { CampaignService } from '../../../services/campaign.service';
 import { CharacterService } from '../../../services/character.service';
 import { NpcService } from '../../../services/npc.service';
+import { RandomTableService } from '../../../services/random-table.service';
 import { PlaythroughService } from '../../../services/playthrough.service';
 import { LayoutService } from '../../../services/layout.service';
 import { PageTitleService } from '../../../services/page-title.service';
@@ -37,6 +38,7 @@ export class PlaythroughFlagsPageComponent implements OnInit, OnDestroy {
     private campaignService: CampaignService,
     private characterService: CharacterService,
     private npcService: NpcService,
+    private randomTableService: RandomTableService,
     private playthroughService: PlaythroughService,
     private layoutService: LayoutService,
     private pageTitleService: PageTitleService
@@ -58,7 +60,7 @@ export class PlaythroughFlagsPageComponent implements OnInit, OnDestroy {
     forkJoin({
       campaign: this.campaignService.getCampaignById(this.campaignId),
       allCampaigns: this.campaignService.getAllCampaigns(),
-      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService),
+      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService, this.randomTableService),
       playthrough: this.playthroughService.getById(this.playthroughId)
     }).subscribe(({ campaign, allCampaigns, treeData, playthrough }) => {
       this.playthrough = playthrough;

web/src/app/campaigns/random-table/random-table-edit/random-table-edit.component.html

@@ -0,0 +1,85 @@
+<div class="rte-page">
+  <div class="rte-toolbar">
+    <button class="btn-back" (click)="back()">
+      <lucide-icon [img]="ArrowLeft" [size]="14"></lucide-icon>
+      Retour
+    </button>
+    <span class="spacer"></span>
+    <button class="btn-save" (click)="save()" [disabled]="saving">
+      <lucide-icon [img]="Save" [size]="14"></lucide-icon>
+      {{ saving ? 'Enregistrement…' : 'Enregistrer' }}
+    </button>
+  </div>
+
+  <h1>{{ tableId ? 'Éditer la table' : 'Nouvelle table aléatoire' }}</h1>
+
+  <div class="error" *ngIf="errorMessage">{{ errorMessage }}</div>
+
+  <div class="form-row">
+    <label for="rt-name">Nom *</label>
+    <input id="rt-name" type="text" [(ngModel)]="name" placeholder="Ex: Rencontres en forêt">
+  </div>
+
+  <div class="form-row">
+    <label for="rt-desc">Description</label>
+    <textarea id="rt-desc" rows="2" [(ngModel)]="description" placeholder="À quoi sert cette table ?"></textarea>
+  </div>
+
+  <div class="form-row">
+    <label for="rt-formula">Formule du dé *</label>
+    <input id="rt-formula" type="text" [(ngModel)]="diceFormula" placeholder="1d20, 2d6, d100…"
+           [class.invalid]="diceFormula && !formulaValid">
+    <small class="hint" [class.bad]="diceFormula && !formulaValid">
+      {{ formulaValid ? 'Formule valide' : 'Format attendu : NdM (ex. 1d20, 2d6, d100)' }}
+    </small>
+  </div>
+
+  <!-- Génération IA -->
+  <div class="ai-box">
+    <div class="ai-title">
+      <lucide-icon [img]="Sparkles" [size]="15"></lucide-icon> Générer avec l'IA
+    </div>
+    <p class="ai-hint">Décris la table ; l'IA propose les entrées (revois-les avant d'enregistrer). Contextualisé avec ta campagne.</p>
+    <textarea rows="2" [(ngModel)]="aiPrompt"
+              placeholder="Ex: rencontres aléatoires dans une forêt hantée, ton sombre"></textarea>
+    <div class="ai-actions">
+      <button class="btn-ai" (click)="generateWithAI()" [disabled]="generating">
+        <lucide-icon [img]="Sparkles" [size]="14"></lucide-icon>
+        {{ generating ? 'Génération…' : 'Générer (' + diceFormula + ')' }}
+      </button>
+      <span class="ai-error" *ngIf="aiError">{{ aiError }}</span>
+    </div>
+  </div>
+
+  <div class="entries-head">
+    <h2>Entrées</h2>
+    <div class="entries-actions">
+      <button class="btn-mini" (click)="autoRanges()" title="Répartir les plages sur la formule">
+        <lucide-icon [img]="Wand2" [size]="13"></lucide-icon> Auto-plages
+      </button>
+      <button class="btn-mini" (click)="addEntry()">
+        <lucide-icon [img]="Plus" [size]="13"></lucide-icon> Ajouter
+      </button>
+    </div>
+  </div>
+
+  <div class="entry-row head">
+    <span class="c-range">Min</span>
+    <span class="c-range">Max</span>
+    <span class="c-label">Résultat</span>
+    <span class="c-detail">Détail</span>
+    <span class="c-del"></span>
+  </div>
+
+  <div class="entry-row" *ngFor="let e of entries; let i = index">
+    <input class="c-range" type="number" [(ngModel)]="e.minRoll">
+    <input class="c-range" type="number" [(ngModel)]="e.maxRoll">
+    <input class="c-label" type="text" [(ngModel)]="e.label" placeholder="Résultat">
+    <input class="c-detail" type="text" [(ngModel)]="e.detail" placeholder="Détail (optionnel)">
+    <button class="c-del btn-del" (click)="removeEntry(i)" title="Supprimer">
+      <lucide-icon [img]="Trash2" [size]="14"></lucide-icon>
+    </button>
+  </div>
+
+  <p class="empty-hint" *ngIf="entries.length === 0">Aucune entrée — clique « Ajouter ».</p>
+</div>

web/src/app/campaigns/random-table/random-table-edit/random-table-edit.component.scss

@@ -0,0 +1,172 @@
+.rte-page {
+  max-width: 900px;
+  margin: 0 auto;
+  padding: 1rem 1.5rem 3rem;
+}
+
+.rte-toolbar {
+  display: flex;
+  align-items: center;
+  gap: 0.5rem;
+  margin-bottom: 1rem;
+
+  .spacer { flex: 1; }
+
+  button {
+    display: inline-flex;
+    align-items: center;
+    gap: 0.35rem;
+    padding: 0.4rem 0.8rem;
+    border-radius: 6px;
+    border: 1px solid rgba(255, 255, 255, 0.12);
+    background: rgba(255, 255, 255, 0.04);
+    color: inherit;
+    cursor: pointer;
+    font-size: 0.85rem;
+    &:hover { background: rgba(255, 255, 255, 0.09); }
+    &:disabled { opacity: 0.5; cursor: default; }
+  }
+  .btn-save { background: #667eea; border-color: #667eea; color: #fff; }
+  .btn-save:hover:not(:disabled) { background: #5568d3; }
+}
+
+h1 { font-size: 1.4rem; margin: 0 0 1rem; }
+h2 { font-size: 1rem; margin: 0; }
+
+.error {
+  padding: 0.6rem 0.9rem;
+  border-radius: 6px;
+  background: rgba(224, 90, 90, 0.12);
+  border: 1px solid rgba(224, 90, 90, 0.4);
+  color: #e88;
+  margin-bottom: 1rem;
+}
+
+.form-row {
+  display: flex;
+  flex-direction: column;
+  gap: 0.3rem;
+  margin-bottom: 1rem;
+
+  label { font-size: 0.85rem; color: var(--color-text-muted, #9aa0aa); }
+
+  input, textarea {
+    padding: 0.5rem 0.7rem;
+    border-radius: 6px;
+    border: 1px solid rgba(255, 255, 255, 0.14);
+    background: rgba(255, 255, 255, 0.04);
+    color: inherit;
+    font: inherit;
+    &.invalid { border-color: rgba(224, 90, 90, 0.6); }
+  }
+
+  .hint { font-size: 0.75rem; color: var(--color-text-muted, #9aa0aa); }
+  .hint.bad { color: #e88; }
+}
+
+.entries-head {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  margin: 1.5rem 0 0.5rem;
+
+  .entries-actions { display: flex; gap: 0.4rem; }
+  .btn-mini {
+    display: inline-flex;
+    align-items: center;
+    gap: 0.3rem;
+    padding: 0.3rem 0.6rem;
+    border-radius: 6px;
+    border: 1px solid rgba(255, 255, 255, 0.14);
+    background: rgba(255, 255, 255, 0.04);
+    color: inherit;
+    cursor: pointer;
+    font-size: 0.8rem;
+    &:hover { background: rgba(255, 255, 255, 0.09); }
+  }
+}
+
+.entry-row {
+  display: grid;
+  grid-template-columns: 70px 70px 1.2fr 1.6fr 36px;
+  gap: 0.5rem;
+  align-items: center;
+  margin-bottom: 0.4rem;
+
+  &.head {
+    font-size: 0.72rem;
+    text-transform: uppercase;
+    letter-spacing: 0.04em;
+    color: var(--color-text-muted, #9aa0aa);
+    margin-bottom: 0.3rem;
+  }
+
+  input {
+    padding: 0.4rem 0.55rem;
+    border-radius: 6px;
+    border: 1px solid rgba(255, 255, 255, 0.14);
+    background: rgba(255, 255, 255, 0.04);
+    color: inherit;
+    font: inherit;
+    width: 100%;
+  }
+
+  .btn-del {
+    display: inline-flex;
+    align-items: center;
+    justify-content: center;
+    padding: 0.35rem;
+    border-radius: 6px;
+    border: 1px solid rgba(224, 90, 90, 0.3);
+    background: rgba(224, 90, 90, 0.08);
+    color: #e88;
+    cursor: pointer;
+    &:hover { background: rgba(224, 90, 90, 0.18); }
+  }
+}
+
+.empty-hint { color: var(--color-text-muted, #9aa0aa); font-style: italic; }
+
+.ai-box {
+  margin: 1.25rem 0 0.5rem;
+  padding: 0.85rem 1rem;
+  border-radius: 10px;
+  background: rgba(168, 130, 255, 0.08);
+  border: 1px solid rgba(168, 130, 255, 0.28);
+
+  .ai-title {
+    display: flex;
+    align-items: center;
+    gap: 0.4rem;
+    font-weight: 600;
+    color: #c4a8ff;
+  }
+  .ai-hint { margin: 0.3rem 0 0.5rem; font-size: 0.8rem; color: var(--color-text-muted, #9aa0aa); }
+
+  textarea {
+    width: 100%;
+    padding: 0.5rem 0.7rem;
+    border-radius: 6px;
+    border: 1px solid rgba(255, 255, 255, 0.14);
+    background: rgba(255, 255, 255, 0.04);
+    color: inherit;
+    font: inherit;
+  }
+
+  .ai-actions { display: flex; align-items: center; gap: 0.7rem; margin-top: 0.5rem; }
+  .btn-ai {
+    display: inline-flex;
+    align-items: center;
+    gap: 0.4rem;
+    padding: 0.45rem 0.9rem;
+    border: none;
+    border-radius: 7px;
+    background: #8a6dff;
+    color: #fff;
+    font-weight: 600;
+    cursor: pointer;
+    &:hover:not(:disabled) { background: #7a5cf0; }
+    &:disabled { opacity: 0.55; cursor: default; }
+  }
+  .ai-error { color: #e88; font-size: 0.82rem; }
+}

web/src/app/campaigns/random-table/random-table-edit/random-table-edit.component.ts

@@ -0,0 +1,185 @@
+import { Component, OnInit } from '@angular/core';
+import { CommonModule } from '@angular/common';
+import { FormsModule } from '@angular/forms';
+import { ActivatedRoute, Router } from '@angular/router';
+import { LucideAngularModule, ArrowLeft, Save, Plus, Trash2, Wand2, Sparkles } from 'lucide-angular';
+import { RandomTableService } from '../../../services/random-table.service';
+import { CampaignSidebarService } from '../../../services/campaign-sidebar.service';
+import { RandomTable, RandomTableEntry, RandomTableCreate } from '../../../services/random-table.model';
+import { DiceUtils } from '../../../shared/dice.utils';
+
+/**
+ * Création/édition d'une table aléatoire (formule de dé + entrées par plage).
+ * Routes : /campaigns/:campaignId/random-tables/create
+ *          /campaigns/:campaignId/random-tables/:tableId/edit
+ */
+@Component({
+  selector: 'app-random-table-edit',
+  standalone: true,
+  imports: [CommonModule, FormsModule, LucideAngularModule],
+  templateUrl: './random-table-edit.component.html',
+  styleUrls: ['./random-table-edit.component.scss']
+})
+export class RandomTableEditComponent implements OnInit {
+  readonly ArrowLeft = ArrowLeft;
+  readonly Save = Save;
+  readonly Plus = Plus;
+  readonly Trash2 = Trash2;
+  readonly Wand2 = Wand2;
+  readonly Sparkles = Sparkles;
+
+  campaignId: string | null = null;
+  tableId: string | null = null;
+
+  name = '';
+  description = '';
+  diceFormula = '1d20';
+  entries: RandomTableEntry[] = [];
+
+  saving = false;
+  errorMessage = '';
+
+  // --- Génération IA ---
+  aiPrompt = '';
+  generating = false;
+  aiError = '';
+
+  constructor(
+    private route: ActivatedRoute,
+    private router: Router,
+    private service: RandomTableService,
+    private campaignSidebar: CampaignSidebarService
+  ) {}
+
+  ngOnInit(): void {
+    const params = this.route.snapshot.paramMap;
+    this.campaignId = params.get('campaignId');
+    this.tableId = params.get('tableId');
+    if (this.campaignId) this.campaignSidebar.show(this.campaignId);
+
+    if (this.tableId) {
+      this.service.getById(this.tableId).subscribe({
+        next: (t: RandomTable) => {
+          this.name = t.name;
+          this.description = t.description ?? '';
+          this.diceFormula = t.diceFormula;
+          this.entries = t.entries.map(e => ({ ...e }));
+        },
+        error: () => this.back()
+      });
+    } else {
+      // Démarre avec une entrée vide pour guider.
+      this.addEntry();
+    }
+  }
+
+  get formulaValid(): boolean {
+    return DiceUtils.parse(this.diceFormula) !== null;
+  }
+
+  addEntry(): void {
+    const range = DiceUtils.totalRange(this.diceFormula);
+    const next = this.entries.length ? this.entries[this.entries.length - 1].maxRoll + 1 : (range?.min ?? 1);
+    this.entries.push({ minRoll: next, maxRoll: next, label: '', detail: '' });
+  }
+
+  removeEntry(i: number): void {
+    this.entries.splice(i, 1);
+  }
+
+  /** Répartit équitablement la plage du dé sur toutes les entrées. */
+  autoRanges(): void {
+    const range = DiceUtils.totalRange(this.diceFormula);
+    if (!range || this.entries.length === 0) return;
+    const span = range.max - range.min + 1;
+    const n = this.entries.length;
+    const base = Math.floor(span / n);
+    let cursor = range.min;
+    this.entries.forEach((e, idx) => {
+      const size = idx === n - 1 ? (range.max - cursor + 1) : Math.max(1, base);
+      e.minRoll = cursor;
+      e.maxRoll = Math.min(range.max, cursor + size - 1);
+      cursor = e.maxRoll + 1;
+    });
+  }
+
+  /** Génère la table via l'IA et préremplit le formulaire (l'utilisateur révise avant d'enregistrer). */
+  generateWithAI(): void {
+    if (!this.campaignId) return;
+    if (!this.aiPrompt.trim()) { this.aiError = 'Décris la table à générer.'; return; }
+    if (!this.formulaValid) { this.aiError = 'Choisis d\'abord une formule de dé valide.'; return; }
+    this.generating = true;
+    this.aiError = '';
+    this.service.generate(this.campaignId, this.aiPrompt.trim(), this.diceFormula.trim()).subscribe({
+      next: (t) => {
+        this.generating = false;
+        if (t.name && !this.name.trim()) this.name = t.name;
+        if (t.description) this.description = t.description;
+        this.entries = (t.entries ?? []).map(e => ({ ...e }));
+      },
+      error: (err) => {
+        this.generating = false;
+        this.aiError = err?.error?.message || 'Échec de la génération IA. Réessaie ou reformule.';
+      }
+    });
+  }
+
+  save(): void {
+    if (!this.campaignId) return;
+    if (!this.name.trim()) { this.errorMessage = 'Le nom est requis.'; return; }
+    if (!this.formulaValid) { this.errorMessage = 'Formule de dé invalide (ex. 1d20, 2d6, d100).'; return; }
+    this.saving = true;
+    this.errorMessage = '';
+
+    const cleanEntries = this.entries
+      .filter(e => e.label.trim())
+      .map(e => ({
+        minRoll: e.minRoll,
+        maxRoll: Math.max(e.minRoll, e.maxRoll),
+        label: e.label.trim(),
+        detail: e.detail?.trim() || undefined
+      }));
+
+    if (this.tableId) {
+      const payload: RandomTable = {
+        id: this.tableId,
+        name: this.name.trim(),
+        description: this.description.trim() || undefined,
+        diceFormula: this.diceFormula.trim(),
+        campaignId: this.campaignId,
+        entries: cleanEntries
+      };
+      this.service.update(this.tableId, payload).subscribe({
+        next: () => this.goToView(this.tableId!),
+        error: (e) => this.fail(e)
+      });
+    } else {
+      const payload: RandomTableCreate = {
+        name: this.name.trim(),
+        description: this.description.trim() || undefined,
+        diceFormula: this.diceFormula.trim(),
+        campaignId: this.campaignId,
+        entries: cleanEntries
+      };
+      this.service.create(payload).subscribe({
+        next: (t) => this.goToView(t.id!),
+        error: (e) => this.fail(e)
+      });
+    }
+  }
+
+  private goToView(id: string): void {
+    this.saving = false;
+    this.router.navigate(['/campaigns', this.campaignId, 'random-tables', id]);
+  }
+
+  private fail(err: unknown): void {
+    this.saving = false;
+    this.errorMessage = 'Échec de l\'enregistrement.';
+    console.error('RandomTable save failed', err);
+  }
+
+  back(): void {
+    this.router.navigate(this.campaignId ? ['/campaigns', this.campaignId] : ['/campaigns']);
+  }
+}

web/src/app/campaigns/random-table/random-table-view/random-table-view.component.html

@@ -0,0 +1,57 @@
+<div class="rt-page" *ngIf="table">
+  <div class="rt-toolbar">
+    <button class="btn-back" (click)="back()">
+      <lucide-icon [img]="ArrowLeft" [size]="14"></lucide-icon>
+      Retour
+    </button>
+    <span class="spacer"></span>
+    <button class="btn-edit" (click)="edit()">
+      <lucide-icon [img]="Edit3" [size]="14"></lucide-icon>
+      Éditer
+    </button>
+  </div>
+
+  <header class="rt-header">
+    <h1><lucide-icon [img]="Dices" [size]="22"></lucide-icon> {{ table.name }}</h1>
+    <p class="rt-desc" *ngIf="table.description">{{ table.description }}</p>
+    <span class="rt-formula">Dé : <code>{{ table.diceFormula }}</code></span>
+  </header>
+
+  <!-- Zone de jet -->
+  <section class="rt-roll">
+    <button class="btn-roll" (click)="roll()">
+      <lucide-icon [img]="Dices" [size]="18"></lucide-icon>
+      Lancer {{ table.diceFormula }}
+    </button>
+    <div class="rt-result" *ngIf="lastRoll">
+      <span class="rt-total">{{ lastRoll.total }}</span>
+      <span class="rt-rolls" *ngIf="lastRoll.rolls.length > 1">({{ lastRoll.rolls.join(' + ') }})</span>
+      <span class="rt-arrow">→</span>
+      <span class="rt-matched" *ngIf="matched">{{ matched.label }}</span>
+      <span class="rt-nomatch" *ngIf="!matched">Aucune entrée pour ce résultat</span>
+    </div>
+  </section>
+
+  <div class="rt-detail" *ngIf="matched?.detail">{{ matched?.detail }}</div>
+
+  <!-- Liste des entrées -->
+  <section class="rt-entries">
+    <div class="rt-empty" *ngIf="table.entries.length === 0">
+      Aucune entrée — édite la table pour en ajouter.
+    </div>
+    <table *ngIf="table.entries.length > 0">
+      <thead>
+        <tr><th class="col-range">Jet</th><th>Résultat</th></tr>
+      </thead>
+      <tbody>
+        <tr *ngFor="let e of table.entries" [class.matched]="isMatched(e)">
+          <td class="col-range">{{ rangeLabel(e) }}</td>
+          <td>
+            <div class="entry-label">{{ e.label }}</div>
+            <div class="entry-detail" *ngIf="e.detail">{{ e.detail }}</div>
+          </td>
+        </tr>
+      </tbody>
+    </table>
+  </section>
+</div>

web/src/app/campaigns/random-table/random-table-view/random-table-view.component.scss

@@ -0,0 +1,133 @@
+.rt-page {
+  max-width: 860px;
+  margin: 0 auto;
+  padding: 1rem 1.5rem 3rem;
+}
+
+.rt-toolbar {
+  display: flex;
+  align-items: center;
+  gap: 0.5rem;
+  margin-bottom: 1rem;
+
+  .spacer { flex: 1; }
+
+  button {
+    display: inline-flex;
+    align-items: center;
+    gap: 0.35rem;
+    padding: 0.4rem 0.7rem;
+    border-radius: 6px;
+    border: 1px solid rgba(255, 255, 255, 0.12);
+    background: rgba(255, 255, 255, 0.04);
+    color: inherit;
+    cursor: pointer;
+    font-size: 0.85rem;
+
+    &:hover { background: rgba(255, 255, 255, 0.09); }
+  }
+}
+
+.rt-header {
+  margin-bottom: 1.25rem;
+
+  h1 {
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+    margin: 0 0 0.35rem;
+    font-size: 1.5rem;
+  }
+
+  .rt-desc { margin: 0 0 0.4rem; color: var(--color-text-muted, #9aa0aa); }
+  .rt-formula code {
+    background: rgba(255, 255, 255, 0.08);
+    padding: 0.1rem 0.4rem;
+    border-radius: 4px;
+  }
+}
+
+.rt-roll {
+  display: flex;
+  align-items: center;
+  gap: 1rem;
+  flex-wrap: wrap;
+  padding: 1rem;
+  border-radius: 10px;
+  background: rgba(102, 126, 234, 0.08);
+  border: 1px solid rgba(102, 126, 234, 0.25);
+  margin-bottom: 1rem;
+
+  .btn-roll {
+    display: inline-flex;
+    align-items: center;
+    gap: 0.5rem;
+    padding: 0.6rem 1.1rem;
+    border: none;
+    border-radius: 8px;
+    background: #667eea;
+    color: #fff;
+    font-weight: 600;
+    cursor: pointer;
+
+    &:hover { background: #5568d3; }
+  }
+
+  .rt-result {
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+    flex-wrap: wrap;
+    font-size: 1.05rem;
+  }
+  .rt-total { font-weight: 700; font-size: 1.3rem; color: #8ea2ff; }
+  .rt-rolls { color: var(--color-text-muted, #9aa0aa); font-size: 0.85rem; }
+  .rt-matched { font-weight: 600; }
+  .rt-nomatch { color: #e0a458; font-style: italic; }
+}
+
+.rt-detail {
+  white-space: pre-wrap;
+  padding: 0.85rem 1rem;
+  margin-bottom: 1.5rem;
+  border-radius: 8px;
+  background: rgba(255, 255, 255, 0.04);
+  border-left: 3px solid #667eea;
+}
+
+.rt-entries {
+  .rt-empty {
+    color: var(--color-text-muted, #9aa0aa);
+    font-style: italic;
+    padding: 1rem 0;
+  }
+
+  table { width: 100%; border-collapse: collapse; }
+
+  th, td {
+    text-align: left;
+    padding: 0.5rem 0.6rem;
+    border-bottom: 1px solid rgba(255, 255, 255, 0.07);
+    vertical-align: top;
+  }
+  th {
+    font-size: 0.75rem;
+    text-transform: uppercase;
+    letter-spacing: 0.04em;
+    color: var(--color-text-muted, #9aa0aa);
+  }
+  .col-range { width: 84px; font-variant-numeric: tabular-nums; white-space: nowrap; }
+
+  .entry-label { font-weight: 500; }
+  .entry-detail {
+    margin-top: 0.2rem;
+    font-size: 0.85rem;
+    color: var(--color-text-muted, #9aa0aa);
+    white-space: pre-wrap;
+  }
+
+  tr.matched {
+    background: rgba(102, 126, 234, 0.16);
+    td { border-bottom-color: rgba(102, 126, 234, 0.3); }
+  }
+}

web/src/app/campaigns/random-table/random-table-view/random-table-view.component.ts

@@ -0,0 +1,85 @@
+import { Component, OnInit } from '@angular/core';
+import { CommonModule } from '@angular/common';
+import { ActivatedRoute, Router } from '@angular/router';
+import { LucideAngularModule, ArrowLeft, Edit3, Dices } from 'lucide-angular';
+import { RandomTableService } from '../../../services/random-table.service';
+import { CampaignSidebarService } from '../../../services/campaign-sidebar.service';
+import { RandomTable, RandomTableEntry } from '../../../services/random-table.model';
+import { DiceUtils, DiceRoll } from '../../../shared/dice.utils';
+
+/**
+ * Vue d'une table aléatoire : affiche les entrées et permet de LANCER le dé
+ * (jet côté client) → surligne l'entrée tombée + montre son détail.
+ * Route : /campaigns/:campaignId/random-tables/:tableId
+ */
+@Component({
+  selector: 'app-random-table-view',
+  standalone: true,
+  imports: [CommonModule, LucideAngularModule],
+  templateUrl: './random-table-view.component.html',
+  styleUrls: ['./random-table-view.component.scss']
+})
+export class RandomTableViewComponent implements OnInit {
+  readonly ArrowLeft = ArrowLeft;
+  readonly Edit3 = Edit3;
+  readonly Dices = Dices;
+
+  campaignId: string | null = null;
+  tableId: string | null = null;
+  table: RandomTable | null = null;
+
+  lastRoll: DiceRoll | null = null;
+  matched: RandomTableEntry | null = null;
+
+  constructor(
+    private route: ActivatedRoute,
+    private router: Router,
+    private service: RandomTableService,
+    private campaignSidebar: CampaignSidebarService
+  ) {}
+
+  ngOnInit(): void {
+    const params = this.route.snapshot.paramMap;
+    this.campaignId = params.get('campaignId');
+    this.tableId = params.get('tableId');
+    if (this.tableId) {
+      this.service.getById(this.tableId).subscribe({
+        next: t => this.table = t,
+        error: () => this.back()
+      });
+    }
+    if (this.campaignId) {
+      this.campaignSidebar.show(this.campaignId);
+    }
+  }
+
+  roll(): void {
+    if (!this.table) return;
+    const result = DiceUtils.roll(this.table.diceFormula);
+    if (!result) { this.lastRoll = null; this.matched = null; return; }
+    this.lastRoll = result;
+    this.matched = this.table.entries.find(
+      e => result.total >= e.minRoll && result.total <= e.maxRoll
+    ) ?? null;
+  }
+
+  isMatched(entry: RandomTableEntry): boolean {
+    return this.matched === entry;
+  }
+
+  rangeLabel(entry: RandomTableEntry): string {
+    return entry.minRoll === entry.maxRoll
+      ? String(entry.minRoll)
+      : `${entry.minRoll}–${entry.maxRoll}`;
+  }
+
+  edit(): void {
+    if (this.campaignId && this.tableId) {
+      this.router.navigate(['/campaigns', this.campaignId, 'random-tables', this.tableId, 'edit']);
+    }
+  }
+
+  back(): void {
+    this.router.navigate(this.campaignId ? ['/campaigns', this.campaignId] : ['/campaigns']);
+  }
+}

web/src/app/campaigns/scene/scene-create/scene-create.component.ts

@@ -7,6 +7,7 @@ import { LucideAngularModule } from 'lucide-angular';
 import { CampaignService } from '../../../services/campaign.service';
 import { CharacterService } from '../../../services/character.service';
 import { NpcService } from '../../../services/npc.service';
+import { RandomTableService } from '../../../services/random-table.service';
 import { LayoutService } from '../../../services/layout.service';
 import { loadCampaignTreeData, buildCampaignSidebarConfig } from '../../campaign-tree.helper';
 import { IconPickerComponent } from '../../../shared/icon-picker/icon-picker.component';
@@ -41,6 +42,7 @@ export class SceneCreateComponent implements OnInit, OnDestroy {
     private campaignService: CampaignService,
     private characterService: CharacterService,
     private npcService: NpcService,
+    private randomTableService: RandomTableService,
     private layoutService: LayoutService
   ) {
     this.form = this.fb.group({
@@ -60,7 +62,7 @@ export class SceneCreateComponent implements OnInit, OnDestroy {
     forkJoin({
       campaign: this.campaignService.getCampaignById(this.campaignId),
       allCampaigns: this.campaignService.getAllCampaigns(),
-      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService)
+      treeData: loadCampaignTreeData(this.campaignService, this.campaignId, this.characterService, this.npcService, this.randomTableService)
     }).subscribe(({ campaign, allCampaigns, treeData }) => {
       const currentChapter = (treeData.chaptersByArc[this.arcId] ?? []).find(c => c.id === this.chapterId);
       this.chapterName = currentChapter?.name ?? '';