Chat persistant pour la partie lore et la partie campagne pour chaque page / scène.....

Correction du carroussel
Passage en v0.4.0
Correction du docker compose pour tout le temps utiliser le bon port que ce soit prod ou dev

brain/app/application/chat.py

@@ -81,6 +81,20 @@ class ChatUseCase:
         ):
             yield token
 
+    def build_system_prompt(
+        self,
+        lore_context: LoreStructuralContext | None = None,
+        page_context: PageContext | None = None,
+        campaign_context: CampaignStructuralContext | None = None,
+        narrative_entity: NarrativeEntityContext | None = None,
+    ) -> str:
+        """Version publique — utilisée par le controller HTTP pour compter
+        les tokens du system prompt avant de streamer (jauge de contexte).
+        """
+        return self._build_system_prompt(
+            lore_context, page_context, campaign_context, narrative_entity
+        )
+
     # --- Construction du system prompt --------------------------------------
 
     def _build_system_prompt(

brain/app/main.py

@@ -9,6 +9,7 @@ from typing import Annotated, AsyncIterator, Literal
 
 import hmac
 import httpx
+import tiktoken
 from fastapi import Depends, FastAPI, HTTPException, Request
 from fastapi.responses import JSONResponse, StreamingResponse
 from pydantic import BaseModel, Field
@@ -37,10 +38,27 @@ from app.infrastructure.onemin_adapter import OneMinAiLLMProvider
 app = FastAPI(
     title="LoreMind Brain",
     description="Backend IA pour la génération de contenu narratif.",
-    version="0.3.0",
+    version="0.4.0",
 )
 
 
+# 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
+# universel approximatif : ±10% d'écart avec Llama/Gemma mais largement
+# suffisant pour une jauge visuelle à l'utilisateur.
+_TOKEN_ENCODER: tiktoken.Encoding | None = None
+
+
+def _count_tokens(text: str | None) -> int:
+    """Compte les tokens d'un texte via tiktoken. Null/empty → 0."""
+    if not text:
+        return 0
+    global _TOKEN_ENCODER
+    if _TOKEN_ENCODER is None:
+        _TOKEN_ENCODER = tiktoken.get_encoding("cl100k_base")
+    return len(_TOKEN_ENCODER.encode(text))
+
+
 # Chemins exemptes d'auth inter-service : healthcheck docker + introspection
 # FastAPI (docs uniquement utiles en dev ; en prod docker-compose, le Brain
 # n'est pas expose en dehors du reseau interne donc pas un risque).
@@ -335,7 +353,32 @@ async def chat_stream(
     campaign_context = _to_campaign_context(body.campaign_context)
     narrative_entity = _to_narrative_entity(body.narrative_entity)
 
+    # --- Comptage tokens pour la jauge de contexte frontend ---
+    # On construit le system prompt une fois ici pour le compter — le use case
+    # le reconstruira à l'identique en interne (coût négligeable : concat de str).
+    # Cette duplication évite de complexifier le contrat stream() avec un
+    # paramètre optionnel system_prompt précalculé.
+    system_prompt_preview = use_case.build_system_prompt(
+        lore_context=lore_context,
+        page_context=page_context,
+        campaign_context=campaign_context,
+        narrative_entity=narrative_entity,
+    )
+    # Dernier message = "current" (souvent user), le reste = historique accumulé.
+    current_msg = messages[-1] if messages else None
+    history_msgs = messages[:-1] if messages else []
+    settings = get_settings()
+    usage_payload = {
+        "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,
+    }
+
     async def event_stream() -> AsyncIterator[str]:
+        # Event 'usage' émis en tout premier : le frontend peut afficher la
+        # jauge avant même le premier token de réponse.
+        yield f"event: usage\ndata: {json.dumps(usage_payload, ensure_ascii=False)}\n\n"
         try:
             async for token in use_case.stream(
                 messages,
@@ -353,6 +396,60 @@ async def chat_stream(
     return StreamingResponse(event_stream(), media_type="text/event-stream")
 
 
+# --- Auto-titre d'une conversation persistee --------------------------------
+
+
+class SummarizeTitleMessageDTO(BaseModel):
+    role: Literal["user", "assistant", "system"]
+    content: str
+
+
+class SummarizeTitleRequestDTO(BaseModel):
+    """Premiers messages d'une conversation pour auto-generer un titre court."""
+
+    messages: list[SummarizeTitleMessageDTO] = Field(default_factory=list)
+
+
+class SummarizeTitleResponseDTO(BaseModel):
+    title: str
+
+
+_TITLE_SYSTEM_PROMPT = (
+    "Tu generes un titre court (4 a 7 mots max) qui resume le sujet de la "
+    "conversation ci-dessous. Reponds UNIQUEMENT par le titre, sans guillemets, "
+    "sans ponctuation finale, sans prefixe type 'Titre :'. Le titre doit etre "
+    "en francais et capturer le sujet metier (pas 'Conversation IA')."
+)
+
+
+@app.post("/summarize/conversation-title", response_model=SummarizeTitleResponseDTO)
+async def summarize_conversation_title(
+    body: SummarizeTitleRequestDTO,
+    llm: Annotated[LLMProvider, Depends(get_llm_provider)],
+) -> SummarizeTitleResponseDTO:
+    """Genere un titre court a partir des premiers echanges de la conversation.
+
+    Appele par le core apres le 1er couple user/assistant, pour remplacer le
+    titre provisoire "Nouvelle conversation" par quelque chose de parlant.
+    """
+    if not body.messages:
+        raise HTTPException(status_code=422, detail="Au moins un message requis")
+
+    transcript = "\n".join(f"{m.role.upper()}: {m.content}" for m in body.messages[:6])
+    prompt = f"{_TITLE_SYSTEM_PROMPT}\n\nConversation :\n{transcript}\n\nTitre :"
+    try:
+        raw = await llm.generate(prompt)
+    except LLMProviderError as exc:
+        raise HTTPException(status_code=502, detail=str(exc)) from exc
+
+    title = raw.strip().splitlines()[0].strip().strip('"').strip("'").rstrip(".")
+    if len(title) > 80:
+        title = title[:80].rstrip()
+    if not title:
+        title = "Nouvelle conversation"
+    return SummarizeTitleResponseDTO(title=title)
+
+
 # --- Mapping DTO → domaine (frontière HTTP) ---------------------------------
 
 
@@ -449,6 +546,9 @@ class SettingsDTO(BaseModel):
     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
+    # Fenetre de contexte effective passee au modele (num_ctx Ollama) — sert
+    # aussi de plafond a la jauge de contexte UI.
+    llm_num_ctx: int
 
 
 class SettingsUpdateDTO(BaseModel):
@@ -460,6 +560,7 @@ class SettingsUpdateDTO(BaseModel):
     onemin_model: str | None = None
     # Chaine vide => on efface la cle. None => pas de changement.
     onemin_api_key: str | None = None
+    llm_num_ctx: int | None = None
 
 
 def _to_settings_dto(s: Settings) -> SettingsDTO:
@@ -469,6 +570,7 @@ 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),
+        llm_num_ctx=s.llm_num_ctx,
     )
 
 
@@ -512,6 +614,50 @@ async def list_ollama_models(
     return {"models": sorted(models)}
 
 
+class OllamaModelInfoDTO(BaseModel):
+    """Info utile extraite de /api/show pour un modele Ollama donne.
+
+    `context_length` = fenetre de contexte max supportee par le modele
+    (extraite des metadonnees GGUF). 0 si inconnue. Le frontend s'en sert
+    pour borner le slider de num_ctx dans les Parametres.
+    """
+
+    context_length: int = 0
+
+
+@app.post("/models/ollama/info", response_model=OllamaModelInfoDTO)
+async def get_ollama_model_info(
+    body: dict[str, str],
+    settings: Annotated[Settings, Depends(get_settings)],
+) -> OllamaModelInfoDTO:
+    """Retourne les metadonnees d'un modele Ollama via /api/show.
+
+    On passe par POST (et pas GET /models/ollama/{name}) parce que les noms
+    Ollama contiennent souvent un `:` (ex: `gemma3:e2b`) qui se segmente
+    mal dans une URL — le body JSON evite le probleme d'escaping.
+
+    Le champ qui nous interesse est `model_info["<arch>.context_length"]`
+    (ex: `gemma3.context_length: 131072`). L'arch varie selon le modele, on
+    scanne donc tous les champs finissant par `.context_length`.
+    """
+    name = (body.get("name") or "").strip()
+    if not name:
+        raise HTTPException(status_code=400, detail="name requis")
+    url = f"{settings.ollama_base_url}/api/show"
+    try:
+        async with httpx.AsyncClient(timeout=5) as client:
+            response = await client.post(url, json={"model": name})
+            response.raise_for_status()
+            data = response.json()
+    except httpx.HTTPError:
+        return OllamaModelInfoDTO(context_length=0)
+    model_info = data.get("model_info") or {}
+    for key, value in model_info.items():
+        if key.endswith(".context_length") and isinstance(value, int):
+            return OllamaModelInfoDTO(context_length=value)
+    return OllamaModelInfoDTO(context_length=0)
+
+
 @app.get("/models/onemin")
 def list_onemin_models() -> dict[str, list[dict[str, object]]]:
     """Catalogue statique des modeles 1min.ai, groupes par fournisseur.

brain/requirements.txt

@@ -3,4 +3,10 @@ uvicorn[standard]==0.32.*
 httpx==0.27.*
 pydantic-settings==2.6.*
 
-pydantic
+pydantic
+
+# Comptage de tokens pour la jauge de contexte (UI chat drawer).
+# L'encodage cl100k_base (GPT-4/3.5) donne une approximation correcte pour
+# la plupart des modeles Llama/Gemma/Mistral (~5-10% d'ecart) — suffisant
+# pour une jauge visuelle.
+tiktoken==0.8.*

core/pom.xml

@@ -14,7 +14,7 @@
 
     <groupId>com.loremind</groupId>
     <artifactId>loremind-core</artifactId>
-    <version>0.3.0</version>
+    <version>0.4.0</version>
     <name>LoreMind Core</name>
     <description>Backend Core - Architecture Hexagonale</description>
 

core/src/main/java/com/loremind/application/conversationcontext/ConversationService.java

@@ -0,0 +1,134 @@
+package com.loremind.application.conversationcontext;
+
+import com.loremind.domain.conversationcontext.Conversation;
+import com.loremind.domain.conversationcontext.ConversationMessage;
+import com.loremind.domain.conversationcontext.ports.ConversationRepository;
+import com.loremind.domain.conversationcontext.ports.ConversationTitleGenerator;
+import org.springframework.stereotype.Service;
+
+import java.util.List;
+import java.util.Optional;
+
+/**
+ * Service d'application du contexte Conversation.
+ *
+ * Regroupe les cas d'usage CRUD + append message + rename. Un seul
+ * service suffit — le contexte est simple et les operations fortement
+ * liees (meme aggregat).
+ *
+ * Regles metier :
+ *  - exactement un ancrage parent (loreId XOR campaignId) ;
+ *  - entityType et entityId vont ensemble (tous deux null = niveau racine,
+ *    tous deux non-null = niveau entite precise).
+ */
+@Service
+public class ConversationService {
+
+    private final ConversationRepository repository;
+    private final ConversationTitleGenerator titleGenerator;
+
+    public ConversationService(ConversationRepository repository,
+                               ConversationTitleGenerator titleGenerator) {
+        this.repository = repository;
+        this.titleGenerator = titleGenerator;
+    }
+
+    /** Donnees de creation d'une conversation. Titre optionnel — sera auto-genere si absent. */
+    public record CreateData(
+            String title,
+            String loreId,
+            String campaignId,
+            String entityType,
+            String entityId) {}
+
+    public Conversation create(CreateData data) {
+        validateAnchor(data.loreId(), data.campaignId(), data.entityType(), data.entityId());
+
+        String title = (data.title() == null || data.title().isBlank())
+                ? "Nouvelle conversation"
+                : data.title().trim();
+
+        Conversation conv = Conversation.builder()
+                .title(title)
+                .loreId(data.loreId())
+                .campaignId(data.campaignId())
+                .entityType(data.entityType())
+                .entityId(data.entityId())
+                .build();
+        return repository.save(conv);
+    }
+
+    public Optional<Conversation> getById(String id) {
+        return repository.findById(id);
+    }
+
+    public List<Conversation> listByContext(String loreId, String campaignId, String entityType, String entityId) {
+        validateAnchor(loreId, campaignId, entityType, entityId);
+        return repository.findByContext(loreId, campaignId, entityType, entityId);
+    }
+
+    public void rename(String id, String title) {
+        if (title == null || title.isBlank()) {
+            throw new IllegalArgumentException("Le titre ne peut pas etre vide");
+        }
+        if (repository.findById(id).isEmpty()) {
+            throw new IllegalArgumentException("Conversation introuvable : " + id);
+        }
+        repository.updateTitle(id, title.trim());
+    }
+
+    public void delete(String id) {
+        repository.deleteById(id);
+    }
+
+    /**
+     * Auto-genere un titre a partir des premiers messages et le persiste.
+     * Appele typiquement apres le 1er couple user/assistant pour remplacer
+     * le titre provisoire. Echec silencieux (fallback dans l'adaptateur) —
+     * on n'empeche pas la conversation de fonctionner si le Brain est down.
+     */
+    public String autoGenerateTitle(String conversationId) {
+        Conversation conv = repository.findById(conversationId)
+                .orElseThrow(() -> new IllegalArgumentException("Conversation introuvable : " + conversationId));
+        List<ConversationMessage> seeds = conv.getMessages();
+        if (seeds == null || seeds.isEmpty()) {
+            return conv.getTitle();
+        }
+        String title = titleGenerator.generate(seeds);
+        repository.updateTitle(conversationId, title);
+        return title;
+    }
+
+    /**
+     * Ajoute un message (user ou assistant) a une conversation existante.
+     * L'horodatage et l'id sont assignes par la couche persistance.
+     */
+    public ConversationMessage appendMessage(String conversationId, String role, String content) {
+        if (role == null || (!role.equals("user") && !role.equals("assistant") && !role.equals("system"))) {
+            throw new IllegalArgumentException("Role invalide : " + role);
+        }
+        if (content == null || content.isEmpty()) {
+            throw new IllegalArgumentException("Contenu vide interdit");
+        }
+        ConversationMessage msg = ConversationMessage.builder()
+                .role(role)
+                .content(content)
+                .build();
+        return repository.appendMessage(conversationId, msg);
+    }
+
+    // ---------- Validation ----------
+
+    private void validateAnchor(String loreId, String campaignId, String entityType, String entityId) {
+        boolean hasLore = loreId != null && !loreId.isBlank();
+        boolean hasCamp = campaignId != null && !campaignId.isBlank();
+        if (hasLore == hasCamp) {
+            throw new IllegalArgumentException("Exactement un parent attendu : loreId XOR campaignId");
+        }
+        boolean hasType = entityType != null && !entityType.isBlank();
+        boolean hasEntId = entityId != null && !entityId.isBlank();
+        if (hasType != hasEntId) {
+            throw new IllegalArgumentException("entityType et entityId doivent etre tous deux null ou tous deux non-null");
+        }
+    }
+}

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

@@ -5,6 +5,7 @@ import com.loremind.domain.campaigncontext.ports.CampaignRepository;
 import com.loremind.domain.generationcontext.CampaignStructuralContext;
 import com.loremind.domain.generationcontext.ChatMessage;
 import com.loremind.domain.generationcontext.ChatRequest;
+import com.loremind.domain.generationcontext.ChatUsage;
 import com.loremind.domain.generationcontext.LoreStructuralContext;
 import com.loremind.domain.generationcontext.NarrativeEntityContext;
 import com.loremind.domain.generationcontext.ports.AiChatProvider;
@@ -65,6 +66,7 @@ public class StreamChatForCampaignUseCase {
             String entityType,
             String entityId,
             List<ChatMessage> messages,
+            Consumer<ChatUsage> onUsage,
             Consumer<String> onToken,
             Runnable onComplete,
             Consumer<Throwable> onError) {
@@ -84,7 +86,7 @@ public class StreamChatForCampaignUseCase {
                 .narrativeEntity(narrativeEntity)
                 .build();
 
-        aiChatProvider.streamChat(request, onToken, onComplete, onError);
+        aiChatProvider.streamChat(request, onUsage, onToken, onComplete, onError);
     }
 
     /**

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

@@ -2,6 +2,7 @@ package com.loremind.application.generationcontext;
 
 import com.loremind.domain.generationcontext.ChatMessage;
 import com.loremind.domain.generationcontext.ChatRequest;
+import com.loremind.domain.generationcontext.ChatUsage;
 import com.loremind.domain.generationcontext.LoreStructuralContext;
 import com.loremind.domain.generationcontext.PageContext;
 import com.loremind.domain.generationcontext.ports.AiChatProvider;
@@ -60,6 +61,7 @@ public class StreamChatForLoreUseCase {
             String loreId,
             String pageId,
             List<ChatMessage> messages,
+            Consumer<ChatUsage> onUsage,
             Consumer<String> onToken,
             Runnable onComplete,
             Consumer<Throwable> onError) {
@@ -75,7 +77,7 @@ public class StreamChatForLoreUseCase {
                 .pageContext(pageContext)
                 .build();
 
-        aiChatProvider.streamChat(request, onToken, onComplete, onError);
+        aiChatProvider.streamChat(request, onUsage, onToken, onComplete, onError);
     }
 
     /**

core/src/main/java/com/loremind/domain/conversationcontext/Conversation.java

@@ -0,0 +1,47 @@
+package com.loremind.domain.conversationcontext;
+
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+import java.time.LocalDateTime;
+import java.util.ArrayList;
+import java.util.List;
+
+/**
+ * Agregat d'une conversation de chat IA persistee.
+ *
+ * Une conversation est ancree sur exactement un niveau de contexte :
+ *  - un Lore (optionnellement une page precise)
+ *  - une Campagne (optionnellement une entite narrative : arc/chapitre/scene)
+ *
+ * C'est cet ancrage qui permet au drawer de filtrer les conversations
+ * a afficher dans la sidebar selon l'ecran en cours.
+ */
+@Data
+@Builder
+@NoArgsConstructor
+@AllArgsConstructor
+public class Conversation {
+
+    private String id;
+    private String title;
+    private LocalDateTime createdAt;
+    private LocalDateTime updatedAt;
+
+    /** Un seul des deux est non-null. */
+    private String loreId;
+    private String campaignId;
+
+    /**
+     * Type d'entite focus, null si la conversation est ancree au niveau
+     * Lore/Campagne racine (pas sur une page/scene precise).
+     * Valeurs : "page", "arc", "chapter", "scene".
+     */
+    private String entityType;
+    private String entityId;
+
+    @Builder.Default
+    private List<ConversationMessage> messages = new ArrayList<>();
+}

core/src/main/java/com/loremind/domain/conversationcontext/ConversationMessage.java

@@ -0,0 +1,28 @@
+package com.loremind.domain.conversationcontext;
+
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+import java.time.LocalDateTime;
+
+/**
+ * Un message persiste d'une conversation.
+ *
+ * Distinct de {@link com.loremind.domain.generationcontext.ChatMessage}
+ * qui reste un simple record role+content pour le streaming LLM. Ici
+ * on ajoute id et horodatage, necessaires pour l'affichage / l'ordre.
+ */
+@Data
+@Builder
+@NoArgsConstructor
+@AllArgsConstructor
+public class ConversationMessage {
+
+    private String id;
+    /** "user" | "assistant" | "system". */
+    private String role;
+    private String content;
+    private LocalDateTime createdAt;
+}

core/src/main/java/com/loremind/domain/conversationcontext/ports/ConversationRepository.java

@@ -0,0 +1,44 @@
+package com.loremind.domain.conversationcontext.ports;
+
+import com.loremind.domain.conversationcontext.Conversation;
+import com.loremind.domain.conversationcontext.ConversationMessage;
+
+import java.util.List;
+import java.util.Optional;
+
+/**
+ * Port de persistance des conversations de chat IA.
+ *
+ * Les methodes de lecture par contexte acceptent des filtres nullables :
+ *  - `loreId` OU `campaignId` doit etre non-null (mais pas les deux)
+ *  - `entityType` + `entityId` : soit tous les deux null (niveau racine),
+ *    soit tous les deux non-null (niveau entite precise).
+ */
+public interface ConversationRepository {
+
+    Conversation save(Conversation conversation);
+
+    Optional<Conversation> findById(String id);
+
+    /**
+     * Liste les conversations filtrees par contexte strict, triees par
+     * updatedAt desc. Les messages ne sont PAS chargees (liste vide) pour
+     * garder la payload legere — la sidebar n'affiche que les titres.
+     */
+    List<Conversation> findByContext(
+            String loreId,
+            String campaignId,
+            String entityType,
+            String entityId);
+
+    void deleteById(String id);
+
+    /**
+     * Ajoute un message a une conversation existante. Met a jour updatedAt
+     * de la conversation parent. Renvoie le message persiste (avec id + ts).
+     */
+    ConversationMessage appendMessage(String conversationId, ConversationMessage message);
+
+    /** Rename atomique — ne touche pas aux messages. */
+    void updateTitle(String conversationId, String title);
+}

core/src/main/java/com/loremind/domain/conversationcontext/ports/ConversationTitleGenerator.java

@@ -0,0 +1,15 @@
+package com.loremind.domain.conversationcontext.ports;
+
+import com.loremind.domain.conversationcontext.ConversationMessage;
+
+import java.util.List;
+
+/**
+ * Port : generation d'un titre court a partir des premiers echanges d'une
+ * conversation. Implemente via un appel Brain /summarize/conversation-title.
+ */
+public interface ConversationTitleGenerator {
+
+    /** Renvoie un titre en francais (4-7 mots max). Jamais null ni vide. */
+    String generate(List<ConversationMessage> firstMessages);
+}

core/src/main/java/com/loremind/domain/generationcontext/ChatUsage.java

@@ -0,0 +1,16 @@
+package com.loremind.domain.generationcontext;
+
+/**
+ * Instantané d'occupation de la fenêtre de contexte à l'instant t du chat.
+ * <p>
+ * Émis une fois par tour de chat (juste avant le streaming des tokens) pour
+ * alimenter la jauge de contexte côté frontend. Les unités sont des tokens
+ * (approximés via tiktoken côté Brain — ±10% vs le tokenizer réel du modèle).
+ *
+ * @param system  tokens consommés par le system prompt (contextes Lore/campagne injectés)
+ * @param history tokens consommés par l'historique de la conversation (hors dernier message)
+ * @param current tokens du dernier message utilisateur en attente de réponse
+ * @param max     taille maximale configurée de la fenêtre de contexte
+ */
+public record ChatUsage(int system, int history, int current, int max) {
+}

core/src/main/java/com/loremind/domain/generationcontext/ports/AiChatProvider.java

@@ -1,6 +1,7 @@
 package com.loremind.domain.generationcontext.ports;
 
 import com.loremind.domain.generationcontext.ChatRequest;
+import com.loremind.domain.generationcontext.ChatUsage;
 
 import java.util.function.Consumer;
 
@@ -26,6 +27,10 @@ public interface AiChatProvider {
      * HTTP côté controller SSE).
      *
      * @param request    messages + contexte Lore
+     * @param onUsage    invoqué une fois au début du stream avec le bilan
+     *                   d'occupation de la fenêtre de contexte (tokens system /
+     *                   history / current / max). Peut ne jamais être invoqué
+     *                   si le provider ne supporte pas le comptage.
      * @param onToken    invoqué à chaque token reçu du LLM (peut être appelé
      *                   de nombreuses fois)
      * @param onComplete invoqué une fois le stream terminé avec succès
@@ -34,6 +39,7 @@ public interface AiChatProvider {
      */
     void streamChat(
             ChatRequest request,
+            Consumer<ChatUsage> onUsage,
             Consumer<String> onToken,
             Runnable onComplete,
             Consumer<Throwable> onError

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

@@ -7,6 +7,7 @@ import com.loremind.domain.generationcontext.CampaignStructuralContext.ChapterSu
 import com.loremind.domain.generationcontext.CampaignStructuralContext.SceneSummary;
 import com.loremind.domain.generationcontext.ChatMessage;
 import com.loremind.domain.generationcontext.ChatRequest;
+import com.loremind.domain.generationcontext.ChatUsage;
 import com.loremind.domain.generationcontext.LoreStructuralContext;
 import com.loremind.domain.generationcontext.LoreStructuralContext.PageSummary;
 import com.loremind.domain.generationcontext.NarrativeEntityContext;
@@ -62,6 +63,7 @@ public class BrainAiChatClient implements AiChatProvider {
     @Override
     public void streamChat(
             ChatRequest request,
+            Consumer<ChatUsage> onUsage,
             Consumer<String> onToken,
             Runnable onComplete,
             Consumer<Throwable> onError) {
@@ -81,7 +83,7 @@ public class BrainAiChatClient implements AiChatProvider {
             // au contrat synchrone du port. L'appelant choisit le thread.
             flux
                 .timeout(Duration.ofSeconds(120))
-                .doOnNext(sse -> handleEvent(sse, onToken, onError))
+                .doOnNext(sse -> handleEvent(sse, onUsage, onToken, onError))
                 .blockLast();
             onComplete.run();
         } catch (Exception e) {
@@ -90,9 +92,10 @@ public class BrainAiChatClient implements AiChatProvider {
         }
     }
 
-    /** Dispatch selon le type d'événement SSE (data par défaut, done, error). */
+    /** Dispatch selon le type d'événement SSE (data par défaut, done, error, usage). */
     private void handleEvent(
             ServerSentEvent<String> sse,
+            Consumer<ChatUsage> onUsage,
             Consumer<String> onToken,
             Consumer<Throwable> onError) {
         String event = sse.event();  // null si pas d'event: xxx -> c'est un data par défaut
@@ -106,6 +109,11 @@ public class BrainAiChatClient implements AiChatProvider {
         if ("done".equals(event)) {
             return; // la fin est gérée par blockLast + onComplete
         }
+        if ("usage".equals(event)) {
+            ChatUsage usage = extractUsage(data);
+            if (usage != null) onUsage.accept(usage);
+            return;
+        }
         // Défaut : événement data avec JSON {"token":"..."}.
         String token = extractToken(data);
         if (token != null && !token.isEmpty()) {
@@ -113,6 +121,39 @@ public class BrainAiChatClient implements AiChatProvider {
         }
     }
 
+    /**
+     * Parse un JSON {"system":N,"history":N,"current":N,"max":N} en ChatUsage.
+     * Renvoie null si le payload est illisible — dans ce cas on ne propage
+     * simplement pas d'usage, le stream token continue normalement.
+     */
+    private ChatUsage extractUsage(String json) {
+        if (json == null) return null;
+        try {
+            int system = extractIntField(json, "system");
+            int history = extractIntField(json, "history");
+            int current = extractIntField(json, "current");
+            int max = extractIntField(json, "max");
+            return new ChatUsage(system, history, current, max);
+        } catch (Exception e) {
+            return null;
+        }
+    }
+
+    /** Parse minimaliste d'un champ entier JSON sans dépendre de Jackson. */
+    private int extractIntField(String json, String field) {
+        String needle = "\"" + field + "\"";
+        int idx = json.indexOf(needle);
+        if (idx < 0) return 0;
+        int colon = json.indexOf(':', idx);
+        if (colon < 0) return 0;
+        int start = colon + 1;
+        while (start < json.length() && Character.isWhitespace(json.charAt(start))) start++;
+        int end = start;
+        while (end < json.length() && (Character.isDigit(json.charAt(end)) || json.charAt(end) == '-')) end++;
+        if (end == start) return 0;
+        return Integer.parseInt(json.substring(start, end));
+    }
+
     /**
      * Parse minimaliste du JSON {"token":"..."} sans pull Jackson ici.
      * Si le format se complexifie, on remplacera par un DTO Jackson.

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

@@ -0,0 +1,68 @@
+package com.loremind.infrastructure.ai;
+
+import com.loremind.domain.conversationcontext.ConversationMessage;
+import com.loremind.domain.conversationcontext.ports.ConversationTitleGenerator;
+import org.springframework.beans.factory.annotation.Value;
+import org.springframework.http.MediaType;
+import org.springframework.stereotype.Component;
+import org.springframework.web.reactive.function.client.WebClient;
+
+import java.time.Duration;
+import java.util.LinkedHashMap;
+import java.util.List;
+import java.util.Map;
+import java.util.stream.Collectors;
+
+/**
+ * Adaptateur : appelle le Brain POST /summarize/conversation-title pour
+ * obtenir un titre court a partir des premiers messages.
+ *
+ * Fallback volontairement silencieux : si le Brain est indisponible, on
+ * renvoie un titre par defaut plutot que de casser l'UX chat.
+ */
+@Component
+public class BrainConversationTitleClient implements ConversationTitleGenerator {
+
+    private static final String PATH = "/summarize/conversation-title";
+    private static final String FALLBACK = "Nouvelle conversation";
+
+    private final WebClient webClient;
+
+    public BrainConversationTitleClient(
+            WebClient.Builder builder,
+            @Value("${brain.base-url}") String baseUrl) {
+        this.webClient = builder.baseUrl(baseUrl).build();
+    }
+
+    @Override
+    public String generate(List<ConversationMessage> firstMessages) {
+        if (firstMessages == null || firstMessages.isEmpty()) {
+            return FALLBACK;
+        }
+        Map<String, Object> payload = new LinkedHashMap<>();
+        payload.put("messages", firstMessages.stream()
+                .map(m -> Map.<String, Object>of(
+                        "role", m.getRole(),
+                        "content", m.getContent() == null ? "" : m.getContent()))
+                .collect(Collectors.toList()));
+
+        try {
+            @SuppressWarnings("unchecked")
+            Map<String, Object> resp = webClient.post()
+                    .uri(PATH)
+                    .contentType(MediaType.APPLICATION_JSON)
+                    .bodyValue(payload)
+                    .retrieve()
+                    .bodyToMono(Map.class)
+                    .timeout(Duration.ofSeconds(20))
+                    .block();
+            if (resp == null) return FALLBACK;
+            Object title = resp.get("title");
+            if (title == null) return FALLBACK;
+            String s = title.toString().trim();
+            return s.isEmpty() ? FALLBACK : s;
+        } catch (Exception e) {
+            return FALLBACK;
+        }
+    }
+}

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

@@ -0,0 +1,89 @@
+package com.loremind.infrastructure.persistence.entity;
+
+import jakarta.persistence.CascadeType;
+import jakarta.persistence.Column;
+import jakarta.persistence.Entity;
+import jakarta.persistence.FetchType;
+import jakarta.persistence.GeneratedValue;
+import jakarta.persistence.GenerationType;
+import jakarta.persistence.Id;
+import jakarta.persistence.Index;
+import jakarta.persistence.OneToMany;
+import jakarta.persistence.OrderBy;
+import jakarta.persistence.PrePersist;
+import jakarta.persistence.PreUpdate;
+import jakarta.persistence.Table;
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+import java.time.LocalDateTime;
+import java.util.ArrayList;
+import java.util.List;
+
+/**
+ * Persistance d'une conversation de chat IA.
+ *
+ * Les refs loreId / campaignId / entityId sont des weak references (String,
+ * pas de FK) — coherent avec la politique inter-contexte du reste du code.
+ * Indexes compose pour accelerer le listing par contexte dans la sidebar.
+ */
+@Entity
+@Table(name = "conversations", indexes = {
+        @Index(name = "idx_conv_lore_entity", columnList = "lore_id,entity_type,entity_id,updated_at"),
+        @Index(name = "idx_conv_campaign_entity", columnList = "campaign_id,entity_type,entity_id,updated_at")
+})
+@Data
+@Builder
+@NoArgsConstructor
+@AllArgsConstructor
+public class ConversationJpaEntity {
+
+    @Id
+    @GeneratedValue(strategy = GenerationType.IDENTITY)
+    private Long id;
+
+    @Column(nullable = false)
+    private String title;
+
+    @Column(name = "lore_id")
+    private String loreId;
+
+    @Column(name = "campaign_id")
+    private String campaignId;
+
+    @Column(name = "entity_type")
+    private String entityType;
+
+    @Column(name = "entity_id")
+    private String entityId;
+
+    @Column(name = "created_at", nullable = false, updatable = false)
+    private LocalDateTime createdAt;
+
+    @Column(name = "updated_at", nullable = false)
+    private LocalDateTime updatedAt;
+
+    /**
+     * Messages enfants. Charges a la demande (fetch=LAZY) pour ne pas plomber
+     * le listing sidebar. Cascade ALL + orphanRemoval : la suppression d'une
+     * conversation efface ses messages.
+     */
+    @OneToMany(mappedBy = "conversation", cascade = CascadeType.ALL, orphanRemoval = true, fetch = FetchType.LAZY)
+    @OrderBy("createdAt ASC, id ASC")
+    @Builder.Default
+    private List<ConversationMessageJpaEntity> messages = new ArrayList<>();
+
+    @PrePersist
+    protected void onCreate() {
+        LocalDateTime now = LocalDateTime.now();
+        createdAt = now;
+        updatedAt = now;
+    }
+
+    @PreUpdate
+    protected void onUpdate() {
+        updatedAt = LocalDateTime.now();
+    }
+}

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

@@ -0,0 +1,59 @@
+package com.loremind.infrastructure.persistence.entity;
+
+import jakarta.persistence.Column;
+import jakarta.persistence.Entity;
+import jakarta.persistence.GeneratedValue;
+import jakarta.persistence.GenerationType;
+import jakarta.persistence.Id;
+import jakarta.persistence.JoinColumn;
+import jakarta.persistence.ManyToOne;
+import jakarta.persistence.PrePersist;
+import jakarta.persistence.Table;
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+import lombok.ToString;
+
+import java.time.LocalDateTime;
+
+/**
+ * Persistance d'un message appartenant a une {@link ConversationJpaEntity}.
+ * Les messages sont ordonnes par createdAt ASC (ordre d'ajout = ordre lu).
+ */
+@Entity
+@Table(name = "conversation_messages")
+@Data
+@Builder
+@NoArgsConstructor
+@AllArgsConstructor
+public class ConversationMessageJpaEntity {
+
+    @Id
+    @GeneratedValue(strategy = GenerationType.IDENTITY)
+    private Long id;
+
+    /** "user" | "assistant" | "system". */
+    @Column(nullable = false, length = 16)
+    private String role;
+
+    @Column(nullable = false, columnDefinition = "TEXT")
+    private String content;
+
+    @Column(name = "created_at", nullable = false, updatable = false)
+    private LocalDateTime createdAt;
+
+    /**
+     * Reference vers la conversation parent. ToString exclu pour eviter une
+     * boucle infinie quand Lombok genere toString() (conv -> messages -> conv...).
+     */
+    @ManyToOne(optional = false)
+    @JoinColumn(name = "conversation_id", nullable = false)
+    @ToString.Exclude
+    private ConversationJpaEntity conversation;
+
+    @PrePersist
+    protected void onCreate() {
+        createdAt = LocalDateTime.now();
+    }
+}

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

@@ -0,0 +1,64 @@
+package com.loremind.infrastructure.persistence.jpa;
+
+import com.loremind.infrastructure.persistence.entity.ConversationJpaEntity;
+import org.springframework.data.jpa.repository.JpaRepository;
+import org.springframework.data.jpa.repository.Query;
+import org.springframework.data.repository.query.Param;
+import org.springframework.stereotype.Repository;
+
+import java.util.List;
+
+/**
+ * Repository Spring Data JPA pour ConversationJpaEntity.
+ *
+ * Les requetes de listing par contexte gerent explicitement les NULL parce
+ * que JPQL `=` ne matche pas NULL. On combine `IS NULL` / `=` selon si le
+ * filtre est fourni — plus simple qu'une Specification Criteria API.
+ */
+@Repository
+public interface ConversationJpaRepository extends JpaRepository<ConversationJpaEntity, Long> {
+
+    /** Listing Lore racine (entity_type IS NULL). */
+    @Query("""
+            SELECT c FROM ConversationJpaEntity c
+            WHERE c.loreId = :loreId
+              AND c.entityType IS NULL
+            ORDER BY c.updatedAt DESC
+            """)
+    List<ConversationJpaEntity> findByLoreRoot(@Param("loreId") String loreId);
+
+    /** Listing Lore + entite precise. */
+    @Query("""
+            SELECT c FROM ConversationJpaEntity c
+            WHERE c.loreId = :loreId
+              AND c.entityType = :entityType
+              AND c.entityId = :entityId
+            ORDER BY c.updatedAt DESC
+            """)
+    List<ConversationJpaEntity> findByLoreAndEntity(
+            @Param("loreId") String loreId,
+            @Param("entityType") String entityType,
+            @Param("entityId") String entityId);
+
+    /** Listing Campagne racine. */
+    @Query("""
+            SELECT c FROM ConversationJpaEntity c
+            WHERE c.campaignId = :campaignId
+              AND c.entityType IS NULL
+            ORDER BY c.updatedAt DESC
+            """)
+    List<ConversationJpaEntity> findByCampaignRoot(@Param("campaignId") String campaignId);
+
+    /** Listing Campagne + entite precise. */
+    @Query("""
+            SELECT c FROM ConversationJpaEntity c
+            WHERE c.campaignId = :campaignId
+              AND c.entityType = :entityType
+              AND c.entityId = :entityId
+            ORDER BY c.updatedAt DESC
+            """)
+    List<ConversationJpaEntity> findByCampaignAndEntity(
+            @Param("campaignId") String campaignId,
+            @Param("entityType") String entityType,
+            @Param("entityId") String entityId);
+}

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

@@ -0,0 +1,148 @@
+package com.loremind.infrastructure.persistence.postgres;
+
+import com.loremind.domain.conversationcontext.Conversation;
+import com.loremind.domain.conversationcontext.ConversationMessage;
+import com.loremind.domain.conversationcontext.ports.ConversationRepository;
+import com.loremind.infrastructure.persistence.entity.ConversationJpaEntity;
+import com.loremind.infrastructure.persistence.entity.ConversationMessageJpaEntity;
+import com.loremind.infrastructure.persistence.jpa.ConversationJpaRepository;
+import jakarta.persistence.EntityNotFoundException;
+import org.springframework.stereotype.Repository;
+import org.springframework.transaction.annotation.Transactional;
+
+import java.util.Collections;
+import java.util.List;
+import java.util.Optional;
+import java.util.stream.Collectors;
+
+/**
+ * Adaptateur Postgres pour ConversationRepository.
+ *
+ * Les methodes de listing ne chargent PAS les messages (messages LAZY,
+ * liste vide renvoyee cote domaine) — la sidebar n'a besoin que des
+ * meta-donnees. findById charge les messages via fetch explicite de la
+ * collection dans une transaction.
+ */
+@Repository
+public class PostgresConversationRepository implements ConversationRepository {
+
+    private final ConversationJpaRepository jpa;
+
+    public PostgresConversationRepository(ConversationJpaRepository jpa) {
+        this.jpa = jpa;
+    }
+
+    @Override
+    @Transactional
+    public Conversation save(Conversation conversation) {
+        ConversationJpaEntity entity = toJpaEntity(conversation);
+        ConversationJpaEntity saved = jpa.save(entity);
+        return toDomain(saved, true);
+    }
+
+    @Override
+    @Transactional(readOnly = true)
+    public Optional<Conversation> findById(String id) {
+        return jpa.findById(Long.parseLong(id))
+                .map(e -> {
+                    // Force l'initialisation LAZY avant de sortir de la transaction.
+                    e.getMessages().size();
+                    return toDomain(e, true);
+                });
+    }
+
+    @Override
+    @Transactional(readOnly = true)
+    public List<Conversation> findByContext(String loreId, String campaignId, String entityType, String entityId) {
+        List<ConversationJpaEntity> rows;
+        if (loreId != null) {
+            rows = (entityType == null)
+                    ? jpa.findByLoreRoot(loreId)
+                    : jpa.findByLoreAndEntity(loreId, entityType, entityId);
+        } else if (campaignId != null) {
+            rows = (entityType == null)
+                    ? jpa.findByCampaignRoot(campaignId)
+                    : jpa.findByCampaignAndEntity(campaignId, entityType, entityId);
+        } else {
+            return Collections.emptyList();
+        }
+        return rows.stream().map(e -> toDomain(e, false)).collect(Collectors.toList());
+    }
+
+    @Override
+    @Transactional
+    public void deleteById(String id) {
+        jpa.deleteById(Long.parseLong(id));
+    }
+
+    @Override
+    @Transactional
+    public ConversationMessage appendMessage(String conversationId, ConversationMessage message) {
+        ConversationJpaEntity conv = jpa.findById(Long.parseLong(conversationId))
+                .orElseThrow(() -> new EntityNotFoundException("Conversation " + conversationId));
+
+        ConversationMessageJpaEntity msg = ConversationMessageJpaEntity.builder()
+                .role(message.getRole())
+                .content(message.getContent())
+                .conversation(conv)
+                .build();
+        conv.getMessages().add(msg);
+        // Force updatedAt via @PreUpdate en modifiant la conv (touch).
+        conv.setUpdatedAt(java.time.LocalDateTime.now());
+
+        ConversationJpaEntity saved = jpa.save(conv);
+        ConversationMessageJpaEntity persisted = saved.getMessages().get(saved.getMessages().size() - 1);
+        return toDomainMessage(persisted);
+    }
+
+    @Override
+    @Transactional
+    public void updateTitle(String conversationId, String title) {
+        ConversationJpaEntity conv = jpa.findById(Long.parseLong(conversationId))
+                .orElseThrow(() -> new EntityNotFoundException("Conversation " + conversationId));
+        conv.setTitle(title);
+        jpa.save(conv);
+    }
+
+    // ---------- Mapping ----------
+
+    private ConversationJpaEntity toJpaEntity(Conversation c) {
+        Long id = c.getId() != null ? Long.parseLong(c.getId()) : null;
+        return ConversationJpaEntity.builder()
+                .id(id)
+                .title(c.getTitle())
+                .loreId(c.getLoreId())
+                .campaignId(c.getCampaignId())
+                .entityType(c.getEntityType())
+                .entityId(c.getEntityId())
+                .createdAt(c.getCreatedAt())
+                .updatedAt(c.getUpdatedAt())
+                .build();
+    }
+
+    private Conversation toDomain(ConversationJpaEntity e, boolean withMessages) {
+        List<ConversationMessage> msgs = withMessages
+                ? e.getMessages().stream().map(this::toDomainMessage).collect(Collectors.toList())
+                : new java.util.ArrayList<>();
+        return Conversation.builder()
+                .id(e.getId().toString())
+                .title(e.getTitle())
+                .loreId(e.getLoreId())
+                .campaignId(e.getCampaignId())
+                .entityType(e.getEntityType())
+                .entityId(e.getEntityId())
+                .createdAt(e.getCreatedAt())
+                .updatedAt(e.getUpdatedAt())
+                .messages(msgs)
+                .build();
+    }
+
+    private ConversationMessage toDomainMessage(ConversationMessageJpaEntity e) {
+        return ConversationMessage.builder()
+                .id(e.getId() != null ? e.getId().toString() : null)
+                .role(e.getRole())
+                .content(e.getContent())
+                .createdAt(e.getCreatedAt())
+                .build();
+    }
+}

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

@@ -3,6 +3,7 @@ package com.loremind.infrastructure.web.controller;
 import com.loremind.application.generationcontext.StreamChatForCampaignUseCase;
 import com.loremind.application.generationcontext.StreamChatForLoreUseCase;
 import com.loremind.domain.generationcontext.ChatMessage;
+import com.loremind.domain.generationcontext.ChatUsage;
 import com.loremind.infrastructure.web.dto.generationcontext.ChatMessageDTO;
 import com.loremind.infrastructure.web.dto.generationcontext.ChatStreamCampaignRequestDTO;
 import com.loremind.infrastructure.web.dto.generationcontext.ChatStreamRequestDTO;
@@ -80,6 +81,7 @@ public class AiChatController {
         try {
             streamChatForLoreUseCase.execute(
                     loreId, pageId, messages,
+                    usage -> sendUsage(emitter, usage),
                     token -> sendToken(emitter, token),
                     () -> complete(emitter),
                     error -> fail(emitter, error));
@@ -100,6 +102,7 @@ public class AiChatController {
         try {
             streamChatForCampaignUseCase.execute(
                     campaignId, entityType, entityId, messages,
+                    usage -> sendUsage(emitter, usage),
                     token -> sendToken(emitter, token),
                     () -> complete(emitter),
                     error -> fail(emitter, error));
@@ -110,6 +113,18 @@ public class AiChatController {
 
     // --- Helpers SSE (un seul point d'écriture par type d'événement) --------
 
+    private void sendUsage(SseEmitter emitter, ChatUsage usage) {
+        try {
+            String payload = "{\"system\":" + usage.system()
+                    + ",\"history\":" + usage.history()
+                    + ",\"current\":" + usage.current()
+                    + ",\"max\":" + usage.max() + "}";
+            emitter.send(SseEmitter.event().name("usage").data(payload));
+        } catch (IOException e) {
+            emitter.completeWithError(e);
+        }
+    }
+
     private void sendToken(SseEmitter emitter, String token) {
         try {
             emitter.send(SseEmitter.event()

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

@@ -0,0 +1,100 @@
+package com.loremind.infrastructure.web.controller;
+
+import com.loremind.application.conversationcontext.ConversationService;
+import com.loremind.domain.conversationcontext.Conversation;
+import com.loremind.domain.conversationcontext.ConversationMessage;
+import com.loremind.infrastructure.web.dto.conversationcontext.AppendMessageDTO;
+import com.loremind.infrastructure.web.dto.conversationcontext.ConversationDTO;
+import com.loremind.infrastructure.web.dto.conversationcontext.ConversationMessageDTO;
+import com.loremind.infrastructure.web.dto.conversationcontext.CreateConversationDTO;
+import com.loremind.infrastructure.web.dto.conversationcontext.RenameConversationDTO;
+import com.loremind.infrastructure.web.mapper.ConversationMapper;
+import org.springframework.http.ResponseEntity;
+import org.springframework.web.bind.annotation.*;
+
+import java.util.List;
+import java.util.stream.Collectors;
+
+/**
+ * API REST des conversations persistees.
+ *
+ * GET /api/conversations?loreId=...&entityType=...&entityId=...   (listing filtre)
+ * GET /api/conversations?campaignId=...&entityType=...&entityId=...
+ * GET /api/conversations/{id}                                     (detail + messages)
+ * POST /api/conversations                                         (create)
+ * PATCH /api/conversations/{id}/title                             (rename)
+ * DELETE /api/conversations/{id}
+ *
+ * L'ajout de messages est piloje cote chat stream (use case dedie),
+ * pas par ce controller.
+ */
+@RestController
+@RequestMapping("/api/conversations")
+public class ConversationController {
+
+    private final ConversationService service;
+    private final ConversationMapper mapper;
+
+    public ConversationController(ConversationService service, ConversationMapper mapper) {
+        this.service = service;
+        this.mapper = mapper;
+    }
+
+    @GetMapping
+    public ResponseEntity<List<ConversationDTO>> list(
+            @RequestParam(required = false) String loreId,
+            @RequestParam(required = false) String campaignId,
+            @RequestParam(required = false) String entityType,
+            @RequestParam(required = false) String entityId) {
+        List<Conversation> rows = service.listByContext(loreId, campaignId, entityType, entityId);
+        return ResponseEntity.ok(rows.stream().map(mapper::toListDTO).collect(Collectors.toList()));
+    }
+
+    @GetMapping("/{id}")
+    public ResponseEntity<ConversationDTO> getById(@PathVariable String id) {
+        return service.getById(id)
+                .map(c -> ResponseEntity.ok(mapper.toDTO(c)))
+                .orElse(ResponseEntity.notFound().build());
+    }
+
+    @PostMapping
+    public ResponseEntity<ConversationDTO> create(@RequestBody CreateConversationDTO dto) {
+        Conversation created = service.create(new ConversationService.CreateData(
+                dto.getTitle(),
+                dto.getLoreId(),
+                dto.getCampaignId(),
+                dto.getEntityType(),
+                dto.getEntityId()));
+        return ResponseEntity.ok(mapper.toDTO(created));
+    }
+
+    @PatchMapping("/{id}/title")
+    public ResponseEntity<Void> rename(@PathVariable String id, @RequestBody RenameConversationDTO dto) {
+        service.rename(id, dto.getTitle());
+        return ResponseEntity.noContent().build();
+    }
+
+    @DeleteMapping("/{id}")
+    public ResponseEntity<Void> delete(@PathVariable String id) {
+        service.delete(id);
+        return ResponseEntity.noContent().build();
+    }
+
+    @PostMapping("/{id}/messages")
+    public ResponseEntity<ConversationMessageDTO> appendMessage(
+            @PathVariable String id,
+            @RequestBody AppendMessageDTO dto) {
+        ConversationMessage saved = service.appendMessage(id, dto.getRole(), dto.getContent());
+        return ResponseEntity.ok(mapper.toMessageDTO(saved));
+    }
+
+    /**
+     * Auto-genere et persiste un titre base sur les premiers messages.
+     * Appele par le front apres le 1er couple user/assistant.
+     */
+    @PostMapping("/{id}/auto-title")
+    public ResponseEntity<RenameConversationDTO> autoTitle(@PathVariable String id) {
+        String title = service.autoGenerateTitle(id);
+        return ResponseEntity.ok(RenameConversationDTO.builder().title(title).build());
+    }
+}

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

@@ -7,6 +7,7 @@ import org.springframework.http.HttpMethod;
 import org.springframework.http.MediaType;
 import org.springframework.http.ResponseEntity;
 import org.springframework.web.bind.annotation.GetMapping;
+import org.springframework.web.bind.annotation.PostMapping;
 import org.springframework.web.bind.annotation.PutMapping;
 import org.springframework.web.bind.annotation.RequestBody;
 import org.springframework.web.bind.annotation.RequestMapping;
@@ -53,6 +54,11 @@ public class SettingsController {
         return forward(HttpMethod.GET, "/models/ollama", null);
     }
 
+    @PostMapping("/models/ollama/info")
+    public ResponseEntity<Map<String, Object>> getOllamaModelInfo(@RequestBody Map<String, Object> body) {
+        return forward(HttpMethod.POST, "/models/ollama/info", body);
+    }
+
     @GetMapping("/models/onemin")
     public ResponseEntity<Map<String, Object>> listOneMinModels() {
         return forward(HttpMethod.GET, "/models/onemin", null);

core/src/main/java/com/loremind/infrastructure/web/dto/conversationcontext/AppendMessageDTO.java

@@ -0,0 +1,16 @@
+package com.loremind.infrastructure.web.dto.conversationcontext;
+
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+@Data
+@Builder
+@NoArgsConstructor
+@AllArgsConstructor
+public class AppendMessageDTO {
+    /** "user" | "assistant" | "system". */
+    private String role;
+    private String content;
+}

core/src/main/java/com/loremind/infrastructure/web/dto/conversationcontext/ConversationDTO.java

@@ -0,0 +1,29 @@
+package com.loremind.infrastructure.web.dto.conversationcontext;
+
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+import java.time.LocalDateTime;
+import java.util.List;
+
+/**
+ * DTO d'une conversation. Les messages sont inclus uniquement sur GET /{id}
+ * (null pour les reponses de listing afin d'alleger la sidebar).
+ */
+@Data
+@Builder
+@NoArgsConstructor
+@AllArgsConstructor
+public class ConversationDTO {
+    private String id;
+    private String title;
+    private String loreId;
+    private String campaignId;
+    private String entityType;
+    private String entityId;
+    private LocalDateTime createdAt;
+    private LocalDateTime updatedAt;
+    private List<ConversationMessageDTO> messages;
+}

core/src/main/java/com/loremind/infrastructure/web/dto/conversationcontext/ConversationMessageDTO.java

@@ -0,0 +1,19 @@
+package com.loremind.infrastructure.web.dto.conversationcontext;
+
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+import java.time.LocalDateTime;
+
+@Data
+@Builder
+@NoArgsConstructor
+@AllArgsConstructor
+public class ConversationMessageDTO {
+    private String id;
+    private String role;
+    private String content;
+    private LocalDateTime createdAt;
+}

core/src/main/java/com/loremind/infrastructure/web/dto/conversationcontext/CreateConversationDTO.java

@@ -0,0 +1,23 @@
+package com.loremind.infrastructure.web.dto.conversationcontext;
+
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+/**
+ * Payload de creation. Le client fournit l'ancrage (lore ou campagne, +/-
+ * entite focus). Le titre est optionnel — sera auto-genere apres le 1er
+ * echange IA si absent.
+ */
+@Data
+@Builder
+@NoArgsConstructor
+@AllArgsConstructor
+public class CreateConversationDTO {
+    private String title;
+    private String loreId;
+    private String campaignId;
+    private String entityType;
+    private String entityId;
+}

core/src/main/java/com/loremind/infrastructure/web/dto/conversationcontext/RenameConversationDTO.java

@@ -0,0 +1,14 @@
+package com.loremind.infrastructure.web.dto.conversationcontext;
+
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+@Data
+@Builder
+@NoArgsConstructor
+@AllArgsConstructor
+public class RenameConversationDTO {
+    private String title;
+}

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

@@ -0,0 +1,61 @@
+package com.loremind.infrastructure.web.mapper;
+
+import com.loremind.domain.conversationcontext.Conversation;
+import com.loremind.domain.conversationcontext.ConversationMessage;
+import com.loremind.infrastructure.web.dto.conversationcontext.ConversationDTO;
+import com.loremind.infrastructure.web.dto.conversationcontext.ConversationMessageDTO;
+import org.springframework.stereotype.Component;
+
+import java.util.List;
+import java.util.stream.Collectors;
+
+/**
+ * Conversion Domaine <-> DTO pour le contexte Conversation.
+ *
+ * {@link #toListDTO(Conversation)} omet les messages — utilise pour le
+ * listing sidebar ou on n'expose que les metadonnees.
+ */
+@Component
+public class ConversationMapper {
+
+    public ConversationDTO toDTO(Conversation c) {
+        List<ConversationMessageDTO> msgs = c.getMessages() == null
+                ? List.of()
+                : c.getMessages().stream().map(this::toMessageDTO).collect(Collectors.toList());
+        return ConversationDTO.builder()
+                .id(c.getId())
+                .title(c.getTitle())
+                .loreId(c.getLoreId())
+                .campaignId(c.getCampaignId())
+                .entityType(c.getEntityType())
+                .entityId(c.getEntityId())
+                .createdAt(c.getCreatedAt())
+                .updatedAt(c.getUpdatedAt())
+                .messages(msgs)
+                .build();
+    }
+
+    /** Variante listing : pas de messages pour alleger la payload. */
+    public ConversationDTO toListDTO(Conversation c) {
+        return ConversationDTO.builder()
+                .id(c.getId())
+                .title(c.getTitle())
+                .loreId(c.getLoreId())
+                .campaignId(c.getCampaignId())
+                .entityType(c.getEntityType())
+                .entityId(c.getEntityId())
+                .createdAt(c.getCreatedAt())
+                .updatedAt(c.getUpdatedAt())
+                .messages(null)
+                .build();
+    }
+
+    public ConversationMessageDTO toMessageDTO(ConversationMessage m) {
+        return ConversationMessageDTO.builder()
+                .id(m.getId())
+                .role(m.getRole())
+                .content(m.getContent())
+                .createdAt(m.getCreatedAt())
+                .build();
+    }
+}

core/src/test/java/com/loremind/application/generationcontext/StreamChatForCampaignUseCaseTest.java

@@ -5,6 +5,7 @@ import com.loremind.domain.campaigncontext.ports.CampaignRepository;
 import com.loremind.domain.generationcontext.CampaignStructuralContext;
 import com.loremind.domain.generationcontext.ChatMessage;
 import com.loremind.domain.generationcontext.ChatRequest;
+import com.loremind.domain.generationcontext.ChatUsage;
 import com.loremind.domain.generationcontext.LoreStructuralContext;
 import com.loremind.domain.generationcontext.NarrativeEntityContext;
 import com.loremind.domain.generationcontext.ports.AiChatProvider;
@@ -46,6 +47,7 @@ public class StreamChatForCampaignUseCaseTest {
 
     private CampaignStructuralContext campaignCtx;
     private List<ChatMessage> messages;
+    private Consumer<ChatUsage> onUsage;
     private Consumer<String> onToken;
     private Runnable onComplete;
     private Consumer<Throwable> onError;
@@ -57,6 +59,7 @@ public class StreamChatForCampaignUseCaseTest {
                 .campaignName("X").campaignDescription("d")
                 .build();
         messages = List.of();
+        onUsage = mock(Consumer.class);
         onToken = mock(Consumer.class);
         onComplete = mock(Runnable.class);
         onError = mock(Consumer.class);
@@ -67,7 +70,7 @@ public class StreamChatForCampaignUseCaseTest {
         when(campaignRepository.findById("missing")).thenReturn(Optional.empty());
 
         assertThrows(IllegalArgumentException.class,
-                () -> useCase.execute("missing", null, null, messages, onToken, onComplete, onError));
+                () -> useCase.execute("missing", null, null, messages, onUsage, onToken, onComplete, onError));
         verifyNoInteractions(aiChatProvider);
     }
 
@@ -77,10 +80,10 @@ public class StreamChatForCampaignUseCaseTest {
         when(campaignRepository.findById("c-1")).thenReturn(Optional.of(standalone));
         when(campaignContextBuilder.build("c-1")).thenReturn(campaignCtx);
 
-        useCase.execute("c-1", null, null, messages, onToken, onComplete, onError);
+        useCase.execute("c-1", null, null, messages, onUsage, onToken, onComplete, onError);
 
         ArgumentCaptor<ChatRequest> captor = ArgumentCaptor.forClass(ChatRequest.class);
-        verify(aiChatProvider).streamChat(captor.capture(), eq(onToken), eq(onComplete), eq(onError));
+        verify(aiChatProvider).streamChat(captor.capture(), eq(onUsage), eq(onToken), eq(onComplete), eq(onError));
         ChatRequest req = captor.getValue();
         assertSame(campaignCtx, req.getCampaignContext());
         assertNull(req.getLoreContext());
@@ -100,10 +103,10 @@ public class StreamChatForCampaignUseCaseTest {
         when(campaignContextBuilder.build("c-1")).thenReturn(campaignCtx);
         when(loreContextBuilder.buildOptional("lore-1")).thenReturn(Optional.of(loreCtx));
 
-        useCase.execute("c-1", null, null, messages, onToken, onComplete, onError);
+        useCase.execute("c-1", null, null, messages, onUsage, onToken, onComplete, onError);
 
         ArgumentCaptor<ChatRequest> captor = ArgumentCaptor.forClass(ChatRequest.class);
-        verify(aiChatProvider).streamChat(captor.capture(), any(), any(), any());
+        verify(aiChatProvider).streamChat(captor.capture(), any(), any(), any(), any());
         assertSame(loreCtx, captor.getValue().getLoreContext());
     }
 
@@ -115,10 +118,10 @@ public class StreamChatForCampaignUseCaseTest {
         when(campaignContextBuilder.build("c-1")).thenReturn(campaignCtx);
         when(loreContextBuilder.buildOptional("lore-ghost")).thenReturn(Optional.empty());
 
-        useCase.execute("c-1", null, null, messages, onToken, onComplete, onError);
+        useCase.execute("c-1", null, null, messages, onUsage, onToken, onComplete, onError);
 
         ArgumentCaptor<ChatRequest> captor = ArgumentCaptor.forClass(ChatRequest.class);
-        verify(aiChatProvider).streamChat(captor.capture(), any(), any(), any());
+        verify(aiChatProvider).streamChat(captor.capture(), any(), any(), any(), any());
         assertNull(captor.getValue().getLoreContext());
         // La requete doit tout de meme partir (pas d'exception).
     }
@@ -133,10 +136,10 @@ public class StreamChatForCampaignUseCaseTest {
         when(campaignContextBuilder.build("c-1")).thenReturn(campaignCtx);
         when(narrativeEntityContextBuilder.build("scene", "s-1")).thenReturn(entity);
 
-        useCase.execute("c-1", "scene", "s-1", messages, onToken, onComplete, onError);
+        useCase.execute("c-1", "scene", "s-1", messages, onUsage, onToken, onComplete, onError);
 
         ArgumentCaptor<ChatRequest> captor = ArgumentCaptor.forClass(ChatRequest.class);
-        verify(aiChatProvider).streamChat(captor.capture(), any(), any(), any());
+        verify(aiChatProvider).streamChat(captor.capture(), any(), any(), any(), any());
         assertSame(entity, captor.getValue().getNarrativeEntity());
     }
 
@@ -146,10 +149,10 @@ public class StreamChatForCampaignUseCaseTest {
         when(campaignRepository.findById("c-1")).thenReturn(Optional.of(standalone));
         when(campaignContextBuilder.build("c-1")).thenReturn(campaignCtx);
 
-        useCase.execute("c-1", "scene", "   ", messages, onToken, onComplete, onError);
+        useCase.execute("c-1", "scene", "   ", messages, onUsage, onToken, onComplete, onError);
 
         ArgumentCaptor<ChatRequest> captor = ArgumentCaptor.forClass(ChatRequest.class);
-        verify(aiChatProvider).streamChat(captor.capture(), any(), any(), any());
+        verify(aiChatProvider).streamChat(captor.capture(), any(), any(), any(), any());
         assertNull(captor.getValue().getNarrativeEntity());
         verifyNoInteractions(narrativeEntityContextBuilder);
     }

core/src/test/java/com/loremind/application/generationcontext/StreamChatForLoreUseCaseTest.java

@@ -2,6 +2,7 @@ package com.loremind.application.generationcontext;
 
 import com.loremind.domain.generationcontext.ChatMessage;
 import com.loremind.domain.generationcontext.ChatRequest;
+import com.loremind.domain.generationcontext.ChatUsage;
 import com.loremind.domain.generationcontext.LoreStructuralContext;
 import com.loremind.domain.generationcontext.ports.AiChatProvider;
 import com.loremind.domain.lorecontext.FieldType;
@@ -46,6 +47,7 @@ public class StreamChatForLoreUseCaseTest {
 
     private LoreStructuralContext loreCtx;
     private List<ChatMessage> messages;
+    private Consumer<ChatUsage> onUsage;
     private Consumer<String> onToken;
     private Runnable onComplete;
     private Consumer<Throwable> onError;
@@ -58,6 +60,7 @@ public class StreamChatForLoreUseCaseTest {
                 .folders(Collections.emptyMap())
                 .build();
         messages = List.of();
+        onUsage = mock(Consumer.class);
         onToken = mock(Consumer.class);
         onComplete = mock(Runnable.class);
         onError = mock(Consumer.class);
@@ -67,10 +70,10 @@ public class StreamChatForLoreUseCaseTest {
     void testExecute_NoPageId_SendsRequestWithoutPageContext() {
         when(loreContextBuilder.build("lore-1")).thenReturn(loreCtx);
 
-        useCase.execute("lore-1", null, messages, onToken, onComplete, onError);
+        useCase.execute("lore-1", null, messages, onUsage, onToken, onComplete, onError);
 
         ArgumentCaptor<ChatRequest> captor = ArgumentCaptor.forClass(ChatRequest.class);
-        verify(aiChatProvider).streamChat(captor.capture(), eq(onToken), eq(onComplete), eq(onError));
+        verify(aiChatProvider).streamChat(captor.capture(), eq(onUsage), eq(onToken), eq(onComplete), eq(onError));
         ChatRequest req = captor.getValue();
         assertSame(loreCtx, req.getLoreContext());
         assertNull(req.getPageContext());
@@ -81,10 +84,10 @@ public class StreamChatForLoreUseCaseTest {
     void testExecute_BlankPageId_TreatedAsNoPage() {
         when(loreContextBuilder.build("lore-1")).thenReturn(loreCtx);
 
-        useCase.execute("lore-1", "   ", messages, onToken, onComplete, onError);
+        useCase.execute("lore-1", "   ", messages, onUsage, onToken, onComplete, onError);
 
         ArgumentCaptor<ChatRequest> captor = ArgumentCaptor.forClass(ChatRequest.class);
-        verify(aiChatProvider).streamChat(captor.capture(), any(), any(), any());
+        verify(aiChatProvider).streamChat(captor.capture(), any(), any(), any(), any());
         assertNull(captor.getValue().getPageContext());
         verifyNoInteractions(pageRepository);
     }
@@ -108,10 +111,10 @@ public class StreamChatForLoreUseCaseTest {
         when(pageRepository.findById("p-1")).thenReturn(Optional.of(page));
         when(templateRepository.findById("tpl-1")).thenReturn(Optional.of(tpl));
 
-        useCase.execute("lore-1", "p-1", messages, onToken, onComplete, onError);
+        useCase.execute("lore-1", "p-1", messages, onUsage, onToken, onComplete, onError);
 
         ArgumentCaptor<ChatRequest> captor = ArgumentCaptor.forClass(ChatRequest.class);
-        verify(aiChatProvider).streamChat(captor.capture(), any(), any(), any());
+        verify(aiChatProvider).streamChat(captor.capture(), any(), any(), any(), any());
         ChatRequest req = captor.getValue();
         assertNotNull(req.getPageContext());
         assertEquals("Alice", req.getPageContext().getTitle());
@@ -130,10 +133,10 @@ public class StreamChatForLoreUseCaseTest {
         when(loreContextBuilder.build("lore-1")).thenReturn(loreCtx);
         when(pageRepository.findById("p-1")).thenReturn(Optional.of(page));
 
-        useCase.execute("lore-1", "p-1", messages, onToken, onComplete, onError);
+        useCase.execute("lore-1", "p-1", messages, onUsage, onToken, onComplete, onError);
 
         ArgumentCaptor<ChatRequest> captor = ArgumentCaptor.forClass(ChatRequest.class);
-        verify(aiChatProvider).streamChat(captor.capture(), any(), any(), any());
+        verify(aiChatProvider).streamChat(captor.capture(), any(), any(), any(), any());
         var pageCtx = captor.getValue().getPageContext();
         assertNotNull(pageCtx);
         assertEquals("Orphan", pageCtx.getTitle());
@@ -153,10 +156,10 @@ public class StreamChatForLoreUseCaseTest {
         when(pageRepository.findById("p-1")).thenReturn(Optional.of(page));
         when(templateRepository.findById("tpl-ghost")).thenReturn(Optional.empty());
 
-        useCase.execute("lore-1", "p-1", messages, onToken, onComplete, onError);
+        useCase.execute("lore-1", "p-1", messages, onUsage, onToken, onComplete, onError);
 
         ArgumentCaptor<ChatRequest> captor = ArgumentCaptor.forClass(ChatRequest.class);
-        verify(aiChatProvider).streamChat(captor.capture(), any(), any(), any());
+        verify(aiChatProvider).streamChat(captor.capture(), any(), any(), any(), any());
         var pageCtx = captor.getValue().getPageContext();
         assertEquals("?", pageCtx.getTemplateName());
         assertTrue(pageCtx.getTemplateFields().isEmpty());
@@ -168,7 +171,7 @@ public class StreamChatForLoreUseCaseTest {
         when(pageRepository.findById("missing")).thenReturn(Optional.empty());
 
         assertThrows(IllegalArgumentException.class,
-                () -> useCase.execute("lore-1", "missing", messages, onToken, onComplete, onError));
+                () -> useCase.execute("lore-1", "missing", messages, onUsage, onToken, onComplete, onError));
         verifyNoInteractions(aiChatProvider);
     }
 }

docker-compose.yml

@@ -31,6 +31,11 @@ services:
       MINIO_ROOT_PASSWORD: ${MINIO_PASSWORD:-minioadmin}
     volumes:
       - minio-data:/data
+    # Mapping bind sur loopback pour autoriser un core/web lance en local (mode dev)
+    # a atteindre MinIO. Invisible sur le LAN donc non-exploitable depuis l'exterieur.
+    ports:
+      - "127.0.0.1:9000:9000"
+      - "127.0.0.1:9001:9001"
     command: server /data --console-address ":9001"
     healthcheck:
       test: ["CMD", "curl", "-f", "http://localhost:9000/minio/health/live"]

web/package.json

@@ -1,6 +1,6 @@
 {
   "name": "loremind-web",
-  "version": "0.3.0",
+  "version": "0.4.0",
   "description": "LoreMind Frontend - Angular",
   "scripts": {
     "ng": "ng",

web/src/app/lore/page-create/page-create.component.html

@@ -78,6 +78,7 @@
 <app-ai-chat-drawer
   [loreId]="loreId"
   [isOpen]="chatOpen"
+  [persistent]="false"
   [welcomeMessage]="wizardWelcome"
   [systemPromptAddon]="wizardSystemPrompt"
   [quickSuggestions]="wizardSuggestions"

web/src/app/services/ai-chat.service.ts

@@ -16,7 +16,19 @@ export interface ChatMessage {
  * - done  : le stream s'est terminé proprement (l'observable va compléter).
  * - error : une erreur s'est produite côté serveur (l'observable va erreur-compléter).
  */
+/**
+ * Instantané d'occupation de la fenêtre de contexte (émis 1x par tour, avant le streaming).
+ * Les valeurs sont exprimées en tokens (~cl100k_base, ±10% vs tokenizer natif du modèle).
+ */
+export interface ChatUsage {
+  system: number;
+  history: number;
+  current: number;
+  max: number;
+}
+
 export type ChatStreamEvent =
+  | { type: 'usage'; usage: ChatUsage }
   | { type: 'token'; value: string }
   | { type: 'done' }
   | { type: 'error'; message: string };
@@ -128,12 +140,19 @@ export class AiChatService {
 
     const dispatchCurrentEvent = () => {
       const eventName = currentEvent ?? 'message';
+      // DEBUG jauge de contexte — à retirer une fois stabilisé.
+      if (eventName !== 'message') {
+        console.log('[AiChatService] SSE event:', eventName, 'data:', currentData);
+      }
       if (eventName === 'error') {
         const message = this.safeParseMessage(currentData);
         subscriber.error(new Error(message));
       } else if (eventName === 'done') {
         subscriber.next({ type: 'done' });
         subscriber.complete();
+      } else if (eventName === 'usage') {
+        const usage = this.safeParseUsage(currentData);
+        if (usage) subscriber.next({ type: 'usage', usage });
       } else {
         // Événement 'message' (défaut) : JSON {"token": "..."}
         const token = this.safeParseToken(currentData);
@@ -188,6 +207,23 @@ export class AiChatService {
     }
   }
 
+  private safeParseUsage(json: string): ChatUsage | null {
+    try {
+      const obj = JSON.parse(json) as Partial<ChatUsage>;
+      if (
+        typeof obj.system === 'number' &&
+        typeof obj.history === 'number' &&
+        typeof obj.current === 'number' &&
+        typeof obj.max === 'number'
+      ) {
+        return { system: obj.system, history: obj.history, current: obj.current, max: obj.max };
+      }
+      return null;
+    } catch {
+      return null;
+    }
+  }
+
   private safeParseMessage(json: string): string {
     try {
       const obj = JSON.parse(json) as { message?: string };

web/src/app/services/conversation.model.ts

@@ -0,0 +1,35 @@
+export type ConversationRole = 'user' | 'assistant' | 'system';
+
+export interface ConversationMessage {
+  id?: string;
+  role: ConversationRole;
+  content: string;
+  createdAt?: string;
+}
+
+export interface Conversation {
+  id: string;
+  title: string;
+  loreId?: string | null;
+  campaignId?: string | null;
+  entityType?: string | null;
+  entityId?: string | null;
+  createdAt: string;
+  updatedAt: string;
+  messages?: ConversationMessage[];
+}
+
+/**
+ * Filtre strict pour le listing sidebar. Fournir soit loreId soit campaignId.
+ * entityType + entityId vont ensemble — tous deux null = niveau racine.
+ */
+export interface ConversationContext {
+  loreId?: string | null;
+  campaignId?: string | null;
+  entityType?: 'page' | 'arc' | 'chapter' | 'scene' | null;
+  entityId?: string | null;
+}
+
+export interface CreateConversationPayload extends ConversationContext {
+  title?: string;
+}

web/src/app/services/conversation.service.ts

@@ -0,0 +1,64 @@
+import { HttpClient, HttpParams } from '@angular/common/http';
+import { Injectable } from '@angular/core';
+import { Observable } from 'rxjs';
+import {
+  Conversation,
+  ConversationContext,
+  ConversationMessage,
+  CreateConversationPayload,
+} from './conversation.model';
+
+/**
+ * Service HTTP des conversations persistees.
+ *
+ * Le streaming (chat/stream) reste pris en charge par AiChatService. Ce
+ * service ne gere que la persistance (metadonnees + messages) et
+ * l'auto-titre declenche apres le 1er echange.
+ */
+@Injectable({ providedIn: 'root' })
+export class ConversationService {
+  private readonly apiUrl = 'http://localhost:8080/api/conversations';
+
+  constructor(private http: HttpClient) {}
+
+  list(ctx: ConversationContext): Observable<Conversation[]> {
+    let params = new HttpParams();
+    if (ctx.loreId) params = params.set('loreId', ctx.loreId);
+    if (ctx.campaignId) params = params.set('campaignId', ctx.campaignId);
+    if (ctx.entityType) params = params.set('entityType', ctx.entityType);
+    if (ctx.entityId) params = params.set('entityId', ctx.entityId);
+    return this.http.get<Conversation[]>(this.apiUrl, { params });
+  }
+
+  getById(id: string): Observable<Conversation> {
+    return this.http.get<Conversation>(`${this.apiUrl}/${id}`);
+  }
+
+  create(payload: CreateConversationPayload): Observable<Conversation> {
+    return this.http.post<Conversation>(this.apiUrl, payload);
+  }
+
+  rename(id: string, title: string): Observable<void> {
+    return this.http.patch<void>(`${this.apiUrl}/${id}/title`, { title });
+  }
+
+  delete(id: string): Observable<void> {
+    return this.http.delete<void>(`${this.apiUrl}/${id}`);
+  }
+
+  appendMessage(
+    id: string,
+    role: 'user' | 'assistant' | 'system',
+    content: string,
+  ): Observable<ConversationMessage> {
+    return this.http.post<ConversationMessage>(`${this.apiUrl}/${id}/messages`, {
+      role,
+      content,
+    });
+  }
+
+  /** Declenche la generation auto du titre cote Brain. */
+  autoTitle(id: string): Observable<{ title: string }> {
+    return this.http.post<{ title: string }>(`${this.apiUrl}/${id}/auto-title`, {});
+  }
+}

web/src/app/services/settings.service.ts

@@ -12,6 +12,7 @@ export interface AppSettings {
   llm_model: string;
   onemin_model: string;
   onemin_api_key_set: boolean;
+  llm_num_ctx: number;
 }
 
 /**
@@ -24,6 +25,13 @@ export interface AppSettingsUpdate {
   llm_model?: string;
   onemin_model?: string;
   onemin_api_key?: string;
+  llm_num_ctx?: number;
+}
+
+/** Metadonnees d'un modele Ollama (issues de /api/show). */
+export interface OllamaModelInfo {
+  /** Fenetre de contexte max du modele (en tokens). 0 si inconnue. */
+  context_length: number;
 }
 
 @Injectable({ providedIn: 'root' })
@@ -49,6 +57,11 @@ export class SettingsService {
     return this.http.get<{ models: string[] }>(`${this.apiUrl}/models/ollama`, this.authOptions);
   }
 
+  getOllamaModelInfo(name: string): Observable<OllamaModelInfo> {
+    return this.http.post<OllamaModelInfo>(
+      `${this.apiUrl}/models/ollama/info`, { name }, this.authOptions);
+  }
+
   listOneMinModels(): Observable<{ groups: OneMinModelGroup[] }> {
     return this.http.get<{ groups: OneMinModelGroup[] }>(`${this.apiUrl}/models/onemin`, this.authOptions);
   }

web/src/app/settings/settings.component.html

@@ -48,7 +48,7 @@
     <div class="form-row">
       <label for="ollama-model">Modele</label>
       <div class="inline-select">
-        <select id="ollama-model" [(ngModel)]="settings.llm_model">
+        <select id="ollama-model" [(ngModel)]="settings.llm_model" (ngModelChange)="fetchOllamaModelInfo()">
           <option *ngIf="ollamaModels.length === 0" [value]="settings.llm_model">{{ settings.llm_model }}</option>
           <option *ngFor="let m of ollamaModels" [value]="m">{{ m }}</option>
         </select>
@@ -93,6 +93,54 @@
     </div>
   </section>
 
+  <!-- Bloc Fenetre de contexte -->
+  <section class="card" *ngIf="settings">
+    <h2>Fenetre de contexte</h2>
+
+    <!-- Ollama : slider borne par le max du modele -->
+    <div class="form-row" *ngIf="settings.llm_provider === 'ollama'">
+      <label for="llm-num-ctx">
+        Tokens alloues au modele
+        <span class="ctx-value">{{ settings.llm_num_ctx | number }}</span>
+        <span class="ctx-max" *ngIf="ollamaModelMaxContext > 0">
+          / {{ ollamaModelMaxContext | number }} max
+        </span>
+      </label>
+      <input
+        id="llm-num-ctx"
+        type="range"
+        [min]="CTX_MIN"
+        [max]="effectiveMaxContext"
+        step="1024"
+        [(ngModel)]="settings.llm_num_ctx"
+        class="ctx-slider">
+      <p class="hint" *ngIf="ollamaModelMaxContext > 0">
+        Le modele <strong>{{ settings.llm_model }}</strong> accepte jusqu'a
+        {{ ollamaModelMaxContext | number }} tokens. Plus la valeur est elevee, plus
+        l'IA peut tenir d'historique et de contexte — au prix de VRAM et de latence.
+      </p>
+      <p class="hint" *ngIf="ollamaModelMaxContext === 0">
+        Impossible de determiner la fenetre max du modele (Ollama injoignable ou modele
+        inconnu). Slider borne a {{ CTX_FALLBACK_MAX | number }} par securite.
+      </p>
+    </div>
+
+    <!-- 1min.ai : saisie libre (pas d'introspection possible) -->
+    <div class="form-row" *ngIf="settings.llm_provider === 'onemin'">
+      <label for="llm-num-ctx-onemin">Fenetre de contexte (tokens)</label>
+      <input
+        id="llm-num-ctx-onemin"
+        type="number"
+        min="2048"
+        step="1024"
+        [(ngModel)]="settings.llm_num_ctx">
+      <p class="hint">
+        A regler selon la capacite du modele 1min.ai choisi (ex: 128 000 pour gpt-4o,
+        200 000 pour claude-sonnet). Sert de plafond a la jauge de contexte du chat.
+      </p>
+    </div>
+  </section>
+
   <div class="actions" *ngIf="settings">
     <button class="btn-primary" (click)="save()" [disabled]="saving">
       <lucide-icon [img]="Save" [size]="16"></lucide-icon>

web/src/app/settings/settings.component.scss

@@ -136,3 +136,20 @@
 }
 .alert-error   { background: rgba(220, 80, 80, 0.15); color: #ff9b9b; }
 .alert-success { background: rgba(80, 200, 120, 0.15); color: #a2e8b6; }
+
+/* --- Slider fenetre de contexte -------------------------------------- */
+.ctx-value {
+  margin-left: 8px;
+  font-variant-numeric: tabular-nums;
+  color: #a5b4fc;
+  font-weight: 600;
+}
+.ctx-max {
+  color: #9ca3af;
+  font-size: 0.85em;
+  font-variant-numeric: tabular-nums;
+}
+.ctx-slider {
+  width: 100%;
+  accent-color: #6c63ff;
+}

web/src/app/settings/settings.component.ts

@@ -42,6 +42,18 @@ export class SettingsComponent implements OnInit {
   errorMessage = '';
   successMessage = '';
 
+  /**
+   * Fenetre de contexte max supportee par le modele Ollama actuellement
+   * selectionne (extraite des metadonnees GGUF via /api/show). 0 si inconnue
+   * — dans ce cas on laisse un fallback de 131072 cote UI.
+   */
+  ollamaModelMaxContext = 0;
+
+  /** Minimum raisonnable pour num_ctx (defaut Ollama = 2048). */
+  readonly CTX_MIN = 2048;
+  /** Fallback si Ollama ne renvoie pas le context_length (modele exotique). */
+  readonly CTX_FALLBACK_MAX = 131072;
+
   /** Cle 1min.ai saisie — vide = on ne touche pas a la cle persistee. */
   oneminApiKeyInput = '';
   /** True si l'utilisateur a coche "effacer la cle". */
@@ -61,6 +73,7 @@ export class SettingsComponent implements OnInit {
       next: (s) => {
         this.settings = { ...s };
         this.refreshModels();
+        this.fetchOllamaModelInfo();
       },
       error: (err) => this.errorMessage = this.extractError(err, 'Impossible de charger les parametres.')
     });
@@ -99,6 +112,31 @@ export class SettingsComponent implements OnInit {
     return group ? group.models : [];
   }
 
+  /**
+   * Recupere la fenetre max supportee par le modele Ollama selectionne.
+   * Si la valeur courante de num_ctx depasse ce max, on la clamp.
+   */
+  fetchOllamaModelInfo(): void {
+    if (!this.settings || this.settings.llm_provider !== 'ollama') return;
+    const modelName = this.settings.llm_model;
+    if (!modelName) return;
+    this.settingsService.getOllamaModelInfo(modelName).subscribe({
+      next: (info) => {
+        this.ollamaModelMaxContext = info.context_length;
+        const max = this.effectiveMaxContext;
+        if (this.settings && this.settings.llm_num_ctx > max) {
+          this.settings.llm_num_ctx = max;
+        }
+      },
+      error: () => this.ollamaModelMaxContext = 0
+    });
+  }
+
+  /** Max effectif a afficher pour le slider (modele Ollama ou fallback). */
+  get effectiveMaxContext(): number {
+    return this.ollamaModelMaxContext > 0 ? this.ollamaModelMaxContext : this.CTX_FALLBACK_MAX;
+  }
+
   /** Quand on change de fournisseur, bascule automatiquement sur son premier modele. */
   onProviderChange(): void {
     if (!this.settings) return;
@@ -118,7 +156,8 @@ export class SettingsComponent implements OnInit {
       llm_provider: this.settings.llm_provider,
       ollama_base_url: this.settings.ollama_base_url,
       llm_model: this.settings.llm_model,
-      onemin_model: this.settings.onemin_model
+      onemin_model: this.settings.onemin_model,
+      llm_num_ctx: this.settings.llm_num_ctx
     };
     if (this.clearApiKey) {
       patch.onemin_api_key = '';

web/src/app/shared/ai-chat-drawer/ai-chat-drawer.component.html

@@ -1,81 +1,148 @@
-<aside class="drawer" [class.drawer-open]="isOpen" aria-label="Assistant IA">
+<aside class="drawer" [class.drawer-open]="isOpen" [class.with-sidebar]="persistent && sidebarOpen" aria-label="Assistant IA">
 
-  <header class="drawer-header">
-    <h2>Assistant IA</h2>
-    <button type="button" class="close-btn" (click)="onClose()" aria-label="Fermer">
-      <lucide-icon [img]="X" [size]="18"></lucide-icon>
-    </button>
-  </header>
-
-  <div #messagesContainer class="messages">
-    <!-- Message d'accueil (non-stocké dans `messages`, toujours visible tant que la conversation est vide). -->
-    <div class="msg msg-assistant" *ngIf="messages.length === 0 && !currentAssistantText">
-      {{ welcomeMessage }}
-    </div>
-
-    <!-- Historique -->
-    <ng-container *ngFor="let m of messages">
-      <div class="msg" [class.msg-user]="m.role === 'user'" [class.msg-assistant]="m.role === 'assistant'">
-        {{ m.content }}
-      </div>
-    </ng-container>
-
-    <!-- Bulle en cours de streaming -->
-    <div class="msg msg-assistant msg-streaming" *ngIf="currentAssistantText">
-      {{ currentAssistantText }}<span class="caret"></span>
-    </div>
-
-    <!-- Indicateur pendant la phase "en train de réfléchir" (avant le premier token) -->
-    <div class="typing-indicator" *ngIf="isStreaming && !currentAssistantText" aria-live="polite">
-      <span></span><span></span><span></span>
-    </div>
-
-    <!-- Erreur locale au drawer -->
-    <div class="msg msg-error" *ngIf="errorMessage" role="alert">
-      {{ errorMessage }}
-    </div>
-  </div>
-
-  <!-- Action primaire (optionnelle) : ne passe PAS par le chat -->
-  <div class="primary-action" *ngIf="primaryAction">
-    <button
-      type="button"
-      class="primary-btn"
-      (click)="onPrimaryAction()"
-      [disabled]="isStreaming">
-      <lucide-icon [img]="Wand2" [size]="14"></lucide-icon>
-      {{ primaryAction.label }}
-    </button>
-  </div>
-
-  <!-- Suggestions rapides -->
-  <div class="quick-suggestions" *ngIf="quickSuggestions.length">
-    <p class="quick-label">Suggestions rapides :</p>
-    <div class="quick-list">
-      <button
-        type="button"
-        class="quick-btn"
-        *ngFor="let s of quickSuggestions"
-        (click)="useQuickSuggestion(s)"
-        [disabled]="isStreaming">
-        <lucide-icon [img]="Lightbulb" [size]="12"></lucide-icon>
-        {{ s }}
+  <!-- Sidebar conversations (mode persistent uniquement) -->
+  <section class="conv-sidebar" *ngIf="persistent && sidebarOpen" aria-label="Conversations">
+    <div class="conv-sidebar-header">
+      <span class="conv-sidebar-title">Conversations</span>
+      <button type="button" class="conv-new-btn" (click)="startNewConversation()" [disabled]="isStreaming" title="Nouvelle conversation">
+        <lucide-icon [img]="MessageSquarePlus" [size]="16"></lucide-icon>
       </button>
     </div>
-  </div>
+    <ul class="conv-list">
+      <li *ngIf="conversations.length === 0" class="conv-empty">Aucune conversation</li>
+      <li
+        *ngFor="let c of conversations"
+        class="conv-item"
+        [class.active]="c.id === currentConversationId"
+        (click)="selectConversation(c)">
+        <span class="conv-item-title">{{ c.title }}</span>
+        <button
+          type="button"
+          class="conv-item-del"
+          (click)="deleteConversation(c, $event)"
+          [disabled]="isStreaming"
+          title="Supprimer">
+          <lucide-icon [img]="Trash2" [size]="12"></lucide-icon>
+        </button>
+      </li>
+    </ul>
+  </section>
 
-  <!-- Zone de saisie -->
-  <form class="input-row" (ngSubmit)="send()">
-    <input
-      type="text"
-      [(ngModel)]="input"
-      name="chatInput"
-      placeholder="Posez une question..."
-      [disabled]="isStreaming"
-      autocomplete="off" />
-    <button type="submit" class="send-btn" [disabled]="!input.trim() || isStreaming" aria-label="Envoyer">
-      <lucide-icon [img]="Send" [size]="16"></lucide-icon>
-    </button>
-  </form>
+  <section class="conv-main">
 
+    <header class="drawer-header">
+      <button
+        *ngIf="persistent"
+        type="button"
+        class="sidebar-toggle"
+        (click)="toggleSidebar()"
+        [attr.aria-label]="sidebarOpen ? 'Masquer la liste' : 'Afficher la liste'"
+        [title]="sidebarOpen ? 'Masquer la liste' : 'Afficher la liste'">
+        <lucide-icon [img]="sidebarOpen ? PanelLeftClose : PanelLeftOpen" [size]="16"></lucide-icon>
+      </button>
+
+      <div class="header-title-wrap">
+        <ng-container *ngIf="persistent && currentConversationId; else defaultTitle">
+          <ng-container *ngIf="!editingTitle; else editingTpl">
+            <h2 class="header-title" [title]="currentTitle">{{ currentTitle || 'Nouvelle conversation' }}</h2>
+            <button type="button" class="rename-btn" (click)="startRenameTitle()" title="Renommer" aria-label="Renommer">
+              <lucide-icon [img]="Pencil" [size]="12"></lucide-icon>
+            </button>
+          </ng-container>
+          <ng-template #editingTpl>
+            <input
+              class="rename-input"
+              type="text"
+              [(ngModel)]="titleDraft"
+              (keyup.enter)="submitRenameTitle()"
+              (keyup.escape)="cancelRenameTitle()"
+              (blur)="submitRenameTitle()"
+              autofocus />
+          </ng-template>
+        </ng-container>
+        <ng-template #defaultTitle>
+          <h2 class="header-title">Assistant IA</h2>
+        </ng-template>
+      </div>
+
+      <button type="button" class="close-btn" (click)="onClose()" aria-label="Fermer">
+        <lucide-icon [img]="X" [size]="18"></lucide-icon>
+      </button>
+    </header>
+
+    <div class="context-gauge" *ngIf="usage" [attr.data-level]="usageLevel"
+         [attr.title]="'System: ' + usage.system + ' · Historique: ' + usage.history + ' · Courant: ' + usage.current + ' / ' + usage.max + ' tokens'">
+      <div class="gauge-bar">
+        <div class="gauge-fill" [style.width.%]="usagePercent"></div>
+      </div>
+      <div class="gauge-label">
+        <span class="gauge-text">Contexte : {{ usageTotal }} / {{ usage.max }} tokens</span>
+        <span class="gauge-percent">{{ usagePercent }}%</span>
+      </div>
+    </div>
+
+    <div #messagesContainer class="messages">
+      <div class="msg msg-assistant" *ngIf="messages.length === 0 && !currentAssistantText">
+        {{ welcomeMessage }}
+      </div>
+
+      <ng-container *ngFor="let m of messages">
+        <div class="msg" [class.msg-user]="m.role === 'user'" [class.msg-assistant]="m.role === 'assistant'">
+          {{ m.content }}
+        </div>
+      </ng-container>
+
+      <div class="msg msg-assistant msg-streaming" *ngIf="currentAssistantText">
+        {{ currentAssistantText }}<span class="caret"></span>
+      </div>
+
+      <div class="typing-indicator" *ngIf="isStreaming && !currentAssistantText" aria-live="polite">
+        <span></span><span></span><span></span>
+      </div>
+
+      <div class="msg msg-error" *ngIf="errorMessage" role="alert">
+        {{ errorMessage }}
+      </div>
+    </div>
+
+    <div class="primary-action" *ngIf="primaryAction">
+      <button
+        type="button"
+        class="primary-btn"
+        (click)="onPrimaryAction()"
+        [disabled]="isStreaming">
+        <lucide-icon [img]="Wand2" [size]="14"></lucide-icon>
+        {{ primaryAction.label }}
+      </button>
+    </div>
+
+    <div class="quick-suggestions" *ngIf="quickSuggestions.length">
+      <p class="quick-label">Suggestions rapides :</p>
+      <div class="quick-list">
+        <button
+          type="button"
+          class="quick-btn"
+          *ngFor="let s of quickSuggestions"
+          (click)="useQuickSuggestion(s)"
+          [disabled]="isStreaming">
+          <lucide-icon [img]="Lightbulb" [size]="12"></lucide-icon>
+          {{ s }}
+        </button>
+      </div>
+    </div>
+
+    <form class="input-row" (ngSubmit)="send()">
+      <input
+        type="text"
+        [(ngModel)]="input"
+        name="chatInput"
+        placeholder="Posez une question..."
+        [disabled]="isStreaming"
+        autocomplete="off" />
+      <button type="submit" class="send-btn" [disabled]="!input.trim() || isStreaming" aria-label="Envoyer">
+        <lucide-icon [img]="Send" [size]="16"></lucide-icon>
+      </button>
+    </form>
+
+  </section>
 </aside>

web/src/app/shared/ai-chat-drawer/ai-chat-drawer.component.scss

@@ -12,13 +12,173 @@
   background: #0f0f1a;
   border-left: 1px solid #1e1e3a;
   display: flex;
-  flex-direction: column;
+  flex-direction: row;
   transform: translateX(100%);
-  transition: transform 0.25s ease;
+  transition: transform 0.25s ease, width 0.25s ease;
   z-index: 1000;
   box-shadow: -4px 0 20px rgba(0, 0, 0, 0.4);
 }
 
+.drawer.with-sidebar {
+  width: 600px;
+}
+
+.conv-sidebar {
+  width: 220px;
+  flex-shrink: 0;
+  display: flex;
+  flex-direction: column;
+  background: #0b0b15;
+  border-right: 1px solid #1e1e3a;
+}
+
+.conv-sidebar-header {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  padding: 0.75rem 0.9rem;
+  border-bottom: 1px solid #1e1e3a;
+
+  .conv-sidebar-title {
+    font-size: 0.78rem;
+    font-weight: 600;
+    color: #9ca3af;
+    text-transform: uppercase;
+    letter-spacing: 0.05em;
+  }
+  .conv-new-btn {
+    background: transparent;
+    border: none;
+    color: #9ca3af;
+    cursor: pointer;
+    padding: 0.3rem;
+    border-radius: 4px;
+    display: flex;
+
+    &:hover:not(:disabled) {
+      background: #1e1e3a;
+      color: white;
+    }
+    &:disabled { opacity: 0.4; cursor: not-allowed; }
+  }
+}
+
+.conv-list {
+  list-style: none;
+  margin: 0;
+  padding: 0.4rem 0;
+  overflow-y: auto;
+  flex: 1;
+}
+
+.conv-empty {
+  padding: 1rem 0.9rem;
+  font-size: 0.78rem;
+  color: #6b7280;
+  font-style: italic;
+}
+
+.conv-item {
+  display: flex;
+  align-items: center;
+  gap: 0.4rem;
+  padding: 0.5rem 0.9rem;
+  cursor: pointer;
+  font-size: 0.82rem;
+  color: #d1d5db;
+  border-left: 2px solid transparent;
+
+  &:hover {
+    background: #14142a;
+  }
+  &.active {
+    background: #1a1a2e;
+    border-left-color: #6c63ff;
+    color: white;
+  }
+  .conv-item-title {
+    flex: 1;
+    white-space: nowrap;
+    overflow: hidden;
+    text-overflow: ellipsis;
+  }
+  .conv-item-del {
+    background: transparent;
+    border: none;
+    color: #6b7280;
+    cursor: pointer;
+    padding: 0.2rem;
+    border-radius: 3px;
+    display: flex;
+    opacity: 0;
+    transition: opacity 0.15s;
+
+    &:hover:not(:disabled) { color: #f87171; background: #1f0f0f; }
+  }
+  &:hover .conv-item-del { opacity: 1; }
+  &.active .conv-item-del { opacity: 1; }
+}
+
+.conv-main {
+  flex: 1;
+  min-width: 0;
+  display: flex;
+  flex-direction: column;
+}
+
+.sidebar-toggle {
+  background: transparent;
+  border: none;
+  color: #9ca3af;
+  cursor: pointer;
+  padding: 0.3rem;
+  border-radius: 4px;
+  display: flex;
+
+  &:hover { background: #1e1e3a; color: white; }
+}
+
+.header-title-wrap {
+  flex: 1;
+  min-width: 0;
+  display: flex;
+  align-items: center;
+  gap: 0.35rem;
+
+  .header-title {
+    margin: 0;
+    font-size: 0.95rem;
+    font-weight: 600;
+    color: white;
+    white-space: nowrap;
+    overflow: hidden;
+    text-overflow: ellipsis;
+    flex: 1;
+  }
+  .rename-btn {
+    background: transparent;
+    border: none;
+    color: #6b7280;
+    cursor: pointer;
+    padding: 0.2rem;
+    border-radius: 3px;
+    display: flex;
+
+    &:hover { color: white; background: #1e1e3a; }
+  }
+  .rename-input {
+    flex: 1;
+    background: #1a1a2e;
+    border: 1px solid #6c63ff;
+    color: white;
+    padding: 0.3rem 0.5rem;
+    border-radius: 4px;
+    font-size: 0.9rem;
+    font-family: inherit;
+    outline: none;
+  }
+}
+
 .drawer-open {
   transform: translateX(0);
 }
@@ -259,3 +419,44 @@
     }
   }
 }
+
+
+/* --- Jauge de contexte ------------------------------------------------- */
+.context-gauge {
+  padding: 0.5rem 1rem 0.75rem;
+  border-bottom: 1px solid #1e1e3a;
+  background: #141428;
+  display: flex;
+  flex-direction: column;
+  gap: 0.35rem;
+}
+
+.context-gauge .gauge-bar {
+  height: 6px;
+  border-radius: 3px;
+  background: #2a2a45;
+  overflow: hidden;
+}
+
+.context-gauge .gauge-fill {
+  height: 100%;
+  transition: width 0.3s ease, background-color 0.3s ease;
+  border-radius: 3px;
+}
+
+.context-gauge[data-level="low"] .gauge-fill  { background: #10b981; }
+.context-gauge[data-level="mid"] .gauge-fill  { background: #f59e0b; }
+.context-gauge[data-level="high"] .gauge-fill { background: #ef4444; }
+
+.context-gauge .gauge-label {
+  display: flex;
+  justify-content: space-between;
+  font-size: 0.72rem;
+  color: #9ca3af;
+  font-variant-numeric: tabular-nums;
+}
+
+.context-gauge[data-level="high"] .gauge-percent {
+  color: #f87171;
+  font-weight: 600;
+}

web/src/app/shared/ai-chat-drawer/ai-chat-drawer.component.ts

@@ -1,105 +1,246 @@
-import { Component, ElementRef, EventEmitter, Input, Output, ViewChild, OnDestroy } from '@angular/core';
 import { CommonModule } from '@angular/common';
+import { Component, ElementRef, EventEmitter, Input, OnChanges, OnDestroy, Output, SimpleChanges, ViewChild } from '@angular/core';
 import { FormsModule } from '@angular/forms';
-import { LucideAngularModule, X, Send, Sparkles, Lightbulb, Wand2 } from 'lucide-angular';
+import { LucideAngularModule, Lightbulb, MessageSquarePlus, PanelLeftClose, PanelLeftOpen, Pencil, Send, Sparkles, Trash2, Wand2, X } from 'lucide-angular';
 import { Subscription } from 'rxjs';
-import { AiChatService, ChatMessage, NarrativeEntityType } from '../../services/ai-chat.service';
+import { AiChatService, ChatMessage, ChatUsage, NarrativeEntityType } from '../../services/ai-chat.service';
+import { Conversation, ConversationContext } from '../../services/conversation.model';
+import { ConversationService } from '../../services/conversation.service';
 
 /**
  * Action primaire optionnelle rendue en gros bouton au-dessus des suggestions.
- * Utilisée pour les actions "spéciales" qui NE passent PAS par le chat
- * (ex: "Remplir automatiquement tous les champs" → déclenche le one-shot b4).
+ * Utilisee pour les actions "speciales" qui NE passent PAS par le chat
+ * (ex: "Remplir automatiquement tous les champs" → declenche le one-shot b4).
  */
 export interface ChatPrimaryAction {
   label: string;
 }
 
 /**
- * Drawer de chat IA réutilisable — panneau fixe à droite de l'écran.
+ * Drawer de chat IA reutilisable — panneau fixe a droite de l'ecran.
  *
- * Usage minimal :
- *   <app-ai-chat-drawer
- *     [loreId]="loreId"
- *     [isOpen]="chatOpen"
- *     [quickSuggestions]="['Développe l'histoire', ...]"
- *     (close)="chatOpen = false">
- *   </app-ai-chat-drawer>
- *
- * Contrainte de design : conversation éphémère (on perd tout à la fermeture
- * ou à la destruction du composant — choix MVP assumé).
+ * Deux modes :
+ *  - `persistent = true` (defaut) : sidebar + conversations persistees en base,
+ *    filtrees par contexte (loreId/campaignId + optionnellement entityType+Id).
+ *    Les messages sont persistes en base au fil du chat et un titre automatique
+ *    est genere apres le 1er echange.
+ *  - `persistent = false` : mode ephemere (pour le wizard de generation de page,
+ *    ou la conversation n'a aucune valeur au-dela de l'usage immediat).
  */
 @Component({
   selector: 'app-ai-chat-drawer',
   standalone: true,
   imports: [CommonModule, FormsModule, LucideAngularModule],
   templateUrl: './ai-chat-drawer.component.html',
-  styleUrls: ['./ai-chat-drawer.component.scss']
+  styleUrls: ['./ai-chat-drawer.component.scss'],
 })
-export class AiChatDrawerComponent implements OnDestroy {
+export class AiChatDrawerComponent implements OnChanges, OnDestroy {
   readonly X = X;
   readonly Send = Send;
   readonly Sparkles = Sparkles;
   readonly Lightbulb = Lightbulb;
   readonly Wand2 = Wand2;
+  readonly MessageSquarePlus = MessageSquarePlus;
+  readonly PanelLeftClose = PanelLeftClose;
+  readonly PanelLeftOpen = PanelLeftOpen;
+  readonly Pencil = Pencil;
+  readonly Trash2 = Trash2;
 
-  /**
-   * Mode Lore : fournir `loreId` (et optionnellement `pageId`).
-   * Mode Campagne : fournir `campaignId` (et optionnellement `entityType`+`entityId`).
-   * Les deux modes sont exclusifs — si `campaignId` est non-vide, on route
-   * vers l'endpoint Campagne, sinon vers l'endpoint Lore.
-   */
   @Input() loreId = '';
-  /**
-   * Optionnel : ID d'une page précise en cours d'édition. Si fourni, le
-   * backend focalise l'IA sur cette page (template, champs, valeurs) via
-   * un bloc "PAGE EN COURS" dans le system prompt. Sans cet ID, le chat
-   * reste générique au Lore.
-   */
   @Input() pageId: string | null = null;
-
-  /** ID de la Campagne — active le mode chat Campagne si non-vide. */
   @Input() campaignId: string | null = null;
-  /** Optionnel : "arc"|"chapter"|"scene" — focalise l'IA sur une entité narrative. */
   @Input() entityType: NarrativeEntityType | null = null;
-  /** Optionnel : ID de l'entité narrative en cours d'édition. */
   @Input() entityId: string | null = null;
   @Input() isOpen = false;
-  /** Texte accueil affiché au premier ouverture (avant tout échange). */
-  @Input() welcomeMessage = 'Bonjour ! Je peux vous aider à développer cette page. Que souhaitez-vous créer ?';
-  /** Suggestions rapides cliquables en bas (hardcodées par le parent, MVP). */
+  @Input() welcomeMessage = 'Bonjour ! Je peux vous aider a developper cette page. Que souhaitez-vous creer ?';
   @Input() quickSuggestions: string[] = [];
-  /** Action primaire optionnelle (ex: "Remplir automatiquement") — ne passe PAS par le chat. */
   @Input() primaryAction: ChatPrimaryAction | null = null;
-  /**
-   * Instructions système supplémentaires injectées en tête de la conversation
-   * envoyée au backend, INVISIBLES côté UI. Usage : mode wizard, où on veut
-   * contextualiser l'IA (template cible, format JSON attendu) sans polluer
-   * l'historique visuel.
-   */
   @Input() systemPromptAddon: string | null = null;
+  /** Persistance activee ? false = mode wizard ephemere. */
+  @Input() persistent = true;
 
   @Output() close = new EventEmitter<void>();
-  /** Émis au clic sur l'action primaire — le parent gère entièrement (one-shot, etc.). */
   @Output() primaryActionClick = new EventEmitter<void>();
-  /** Émis à chaque fin de réponse assistant — utile pour parser côté parent (ex: bloc <values> du wizard). */
   @Output() assistantReply = new EventEmitter<string>();
 
   @ViewChild('messagesContainer') messagesContainer?: ElementRef<HTMLDivElement>;
 
-  /** Conversation en cours (user + assistant). Le welcome n'est pas dedans — rendu séparément. */
   messages: ChatMessage[] = [];
-  /** Texte en cours de streaming (écrit token par token, pas encore poussé dans `messages`). */
   currentAssistantText = '';
-  /** Champ de saisie. */
   input = '';
-  /** Stream en cours ? Désactive le bouton envoyer + les suggestions rapides. */
   isStreaming = false;
-  /** Dernier message d'erreur (affiché dans une bannière locale au drawer). */
   errorMessage: string | null = null;
+  usage: ChatUsage | null = null;
+
+  // --- Persistance --------------------------------------------------------
+
+  /** Liste visible dans la sidebar pour le contexte courant. */
+  conversations: Conversation[] = [];
+  /** Conversation actuellement chargee (null = nouvelle conversation vierge). */
+  currentConversationId: string | null = null;
+  /** Titre de la conversation courante (affiche dans le header). */
+  currentTitle = '';
+  /** Mode edition inline du titre. */
+  editingTitle = false;
+  titleDraft = '';
+  /** Etat repliable de la sidebar. */
+  sidebarOpen = true;
 
   private streamSub: Subscription | null = null;
 
-  constructor(private readonly chatService: AiChatService) {}
+  constructor(
+    private readonly chatService: AiChatService,
+    private readonly conversationService: ConversationService,
+  ) {}
+
+  // --- Jauge de contexte --------------------------------------------------
+
+  get usageTotal(): number {
+    if (!this.usage) return 0;
+    return this.usage.system + this.usage.history + this.usage.current;
+  }
+  get usageRatio(): number {
+    if (!this.usage || this.usage.max <= 0) return 0;
+    return Math.min(1, this.usageTotal / this.usage.max);
+  }
+  get usagePercent(): number {
+    return Math.round(this.usageRatio * 100);
+  }
+  get usageLevel(): 'low' | 'mid' | 'high' {
+    const r = this.usageRatio;
+    if (r > 0.8) return 'high';
+    if (r >= 0.5) return 'mid';
+    return 'low';
+  }
+
+  // --- Cycle de vie -------------------------------------------------------
+
+  ngOnChanges(changes: SimpleChanges): void {
+    if (!this.persistent) return;
+    const contextChanged =
+      changes['loreId'] || changes['pageId'] || changes['campaignId'] || changes['entityType'] || changes['entityId'];
+    const openedNow = changes['isOpen'] && this.isOpen;
+    if (contextChanged || openedNow) {
+      this.resetConversationState();
+      this.reloadConversations();
+    }
+  }
+
+  ngOnDestroy(): void {
+    this.abortStream();
+  }
+
+  // --- Sidebar : listing / nouveau / select / rename / delete ------------
+
+  private buildContext(): ConversationContext {
+    // Cote Lore : pageId joue le role de focus entite (entityType="page").
+    // Cote Campagne : entityType + entityId sont deja fournis directement.
+    if (this.loreId) {
+      return {
+        loreId: this.loreId,
+        campaignId: null,
+        entityType: this.pageId ? 'page' : null,
+        entityId: this.pageId ?? null,
+      };
+    }
+    return {
+      loreId: null,
+      campaignId: this.campaignId || null,
+      entityType: this.entityType,
+      entityId: this.entityId,
+    };
+  }
+
+  reloadConversations(): void {
+    if (!this.persistent) return;
+    const ctx = this.buildContext();
+    if (!ctx.loreId && !ctx.campaignId) {
+      this.conversations = [];
+      return;
+    }
+    this.conversationService.list(ctx).subscribe({
+      next: (rows) => (this.conversations = rows),
+      error: () => (this.conversations = []),
+    });
+  }
+
+  startNewConversation(): void {
+    if (this.isStreaming) return;
+    this.resetConversationState();
+  }
+
+  private resetConversationState(): void {
+    this.currentConversationId = null;
+    this.currentTitle = '';
+    this.messages = [];
+    this.currentAssistantText = '';
+    this.errorMessage = null;
+    this.usage = null;
+    this.editingTitle = false;
+  }
+
+  selectConversation(conv: Conversation): void {
+    if (this.isStreaming) return;
+    this.conversationService.getById(conv.id).subscribe({
+      next: (full) => {
+        this.currentConversationId = full.id;
+        this.currentTitle = full.title;
+        this.messages = (full.messages ?? [])
+          .filter((m) => m.role === 'user' || m.role === 'assistant')
+          .map((m) => ({ role: m.role as 'user' | 'assistant', content: m.content }));
+        this.currentAssistantText = '';
+        this.errorMessage = null;
+        this.usage = null;
+        this.scrollToBottom();
+      },
+      error: () => (this.errorMessage = 'Impossible de charger la conversation.'),
+    });
+  }
+
+  deleteConversation(conv: Conversation, event: Event): void {
+    event.stopPropagation();
+    if (this.isStreaming) return;
+    if (!confirm(`Supprimer la conversation "${conv.title}" ?`)) return;
+    this.conversationService.delete(conv.id).subscribe({
+      next: () => {
+        this.conversations = this.conversations.filter((c) => c.id !== conv.id);
+        if (this.currentConversationId === conv.id) this.resetConversationState();
+      },
+    });
+  }
+
+  startRenameTitle(): void {
+    if (!this.currentConversationId) return;
+    this.titleDraft = this.currentTitle;
+    this.editingTitle = true;
+  }
+
+  cancelRenameTitle(): void {
+    this.editingTitle = false;
+    this.titleDraft = '';
+  }
+
+  submitRenameTitle(): void {
+    const t = this.titleDraft.trim();
+    if (!t || !this.currentConversationId) {
+      this.cancelRenameTitle();
+      return;
+    }
+    const id = this.currentConversationId;
+    this.conversationService.rename(id, t).subscribe({
+      next: () => {
+        this.currentTitle = t;
+        this.conversations = this.conversations.map((c) =>
+          c.id === id ? { ...c, title: t } : c,
+        );
+        this.editingTitle = false;
+      },
+    });
+  }
+
+  toggleSidebar(): void {
+    this.sidebarOpen = !this.sidebarOpen;
+  }
 
   // --- Handlers UI --------------------------------------------------------
 
@@ -108,7 +249,6 @@ export class AiChatDrawerComponent implements OnDestroy {
     this.close.emit();
   }
 
-  /** Envoi explicite depuis le formulaire (Entrée ou bouton envoyer). */
   send(): void {
     const text = this.input.trim();
     if (!text || this.isStreaming) return;
@@ -116,45 +256,114 @@ export class AiChatDrawerComponent implements OnDestroy {
     this.input = '';
   }
 
-  /** Envoi depuis une suggestion rapide (bouton cliquable en bas). */
   useQuickSuggestion(suggestion: string): void {
     if (this.isStreaming) return;
     this.sendUserMessage(suggestion);
   }
 
-  /** Clic sur l'action primaire — on délègue entièrement au parent. */
   onPrimaryAction(): void {
     if (this.isStreaming) return;
     this.primaryActionClick.emit();
   }
 
-  // --- Logique envoi + streaming -----------------------------------------
+  // --- Envoi + streaming --------------------------------------------------
 
   private sendUserMessage(text: string): void {
+    if (this.persistent) {
+      this.ensureConversation().then((convId) => {
+        if (convId) this.streamAndPersist(text, convId);
+      });
+    } else {
+      this.streamEphemeral(text);
+    }
+  }
+
+  /**
+   * Cree la conversation cote serveur si elle n'existe pas encore. Resolu
+   * avec l'id, ou null sur erreur (auquel cas on n'envoie pas).
+   */
+  private ensureConversation(): Promise<string | null> {
+    if (this.currentConversationId) return Promise.resolve(this.currentConversationId);
+    const ctx = this.buildContext();
+    if (!ctx.loreId && !ctx.campaignId) {
+      this.errorMessage = 'Contexte manquant pour creer une conversation.';
+      return Promise.resolve(null);
+    }
+    return new Promise((resolve) => {
+      this.conversationService.create(ctx).subscribe({
+        next: (conv) => {
+          this.currentConversationId = conv.id;
+          this.currentTitle = conv.title;
+          this.conversations = [conv, ...this.conversations];
+          resolve(conv.id);
+        },
+        error: () => {
+          this.errorMessage = 'Impossible de creer la conversation.';
+          resolve(null);
+        },
+      });
+    });
+  }
+
+  private streamAndPersist(text: string, convId: string): void {
+    const wasEmpty = this.messages.length === 0;
     this.errorMessage = null;
     this.messages.push({ role: 'user', content: text });
     this.currentAssistantText = '';
     this.isStreaming = true;
     this.scrollToBottom();
 
-    // Construit la liste effectivement envoyée au backend : systemPromptAddon
-    // (si fourni) préfixé, puis l'historique visible. Le system n'est PAS stocké
-    // dans this.messages → reste invisible côté UI.
-    const payload = this.systemPromptAddon
-      ? [{ role: 'system' as const, content: this.systemPromptAddon }, ...this.messages]
-      : this.messages;
+    // Persiste le message user immediatement — evite toute perte si stream interrompu.
+    this.conversationService.appendMessage(convId, 'user', text).subscribe({ error: () => {} });
 
-    const stream$ = this.campaignId
-      ? this.chatService.streamChatForCampaign(this.campaignId, payload, this.entityType, this.entityId)
-      : this.chatService.streamChat(this.loreId, payload, this.pageId);
-
-    this.streamSub = stream$.subscribe({
+    this.streamSub = this.buildStream().subscribe({
       next: (event) => {
         if (event.type === 'token') {
           this.currentAssistantText += event.value;
           this.scrollToBottom();
+        } else if (event.type === 'usage') {
+          this.usage = event.usage;
+        }
+      },
+      error: (err) => {
+        this.isStreaming = false;
+        this.errorMessage = err?.message ?? 'Erreur inconnue.';
+        this.currentAssistantText = '';
+      },
+      complete: () => {
+        const reply = this.currentAssistantText;
+        if (reply) {
+          this.messages.push({ role: 'assistant', content: reply });
+          this.assistantReply.emit(reply);
+          this.conversationService.appendMessage(convId, 'assistant', reply).subscribe({
+            next: () => {
+              if (wasEmpty) this.triggerAutoTitle(convId);
+            },
+            error: () => {},
+          });
+        }
+        this.currentAssistantText = '';
+        this.isStreaming = false;
+        this.scrollToBottom();
+      },
+    });
+  }
+
+  private streamEphemeral(text: string): void {
+    this.errorMessage = null;
+    this.messages.push({ role: 'user', content: text });
+    this.currentAssistantText = '';
+    this.isStreaming = true;
+    this.scrollToBottom();
+
+    this.streamSub = this.buildStream().subscribe({
+      next: (event) => {
+        if (event.type === 'token') {
+          this.currentAssistantText += event.value;
+          this.scrollToBottom();
+        } else if (event.type === 'usage') {
+          this.usage = event.usage;
         }
-        // 'done' : l'Observable va compléter → géré par complete()
       },
       error: (err) => {
         this.isStreaming = false;
@@ -162,7 +371,6 @@ export class AiChatDrawerComponent implements OnDestroy {
         this.currentAssistantText = '';
       },
       complete: () => {
-        // On fige le texte streamé en message assistant réel, puis on reset le buffer.
         const reply = this.currentAssistantText;
         if (reply) {
           this.messages.push({ role: 'assistant', content: reply });
@@ -171,7 +379,28 @@ export class AiChatDrawerComponent implements OnDestroy {
         this.currentAssistantText = '';
         this.isStreaming = false;
         this.scrollToBottom();
-      }
+      },
+    });
+  }
+
+  private buildStream() {
+    const payload = this.systemPromptAddon
+      ? [{ role: 'system' as const, content: this.systemPromptAddon }, ...this.messages]
+      : this.messages;
+    return this.campaignId
+      ? this.chatService.streamChatForCampaign(this.campaignId, payload, this.entityType, this.entityId)
+      : this.chatService.streamChat(this.loreId, payload, this.pageId);
+  }
+
+  private triggerAutoTitle(convId: string): void {
+    this.conversationService.autoTitle(convId).subscribe({
+      next: ({ title }) => {
+        this.currentTitle = title;
+        this.conversations = this.conversations.map((c) =>
+          c.id === convId ? { ...c, title } : c,
+        );
+      },
+      error: () => {},
     });
   }
 
@@ -182,18 +411,10 @@ export class AiChatDrawerComponent implements OnDestroy {
     this.currentAssistantText = '';
   }
 
-  /**
-   * Scroll différé au prochain tick : donne à Angular le temps de rendre
-   * le nouveau contenu avant qu'on mesure/ajuste la position du scroll.
-   */
   private scrollToBottom(): void {
     queueMicrotask(() => {
       const el = this.messagesContainer?.nativeElement;
       if (el) el.scrollTop = el.scrollHeight;
     });
   }
-
-  ngOnDestroy(): void {
-    this.abortStream();
-  }
 }

web/src/app/shared/image-gallery/image-gallery.component.scss

@@ -163,20 +163,6 @@
   align-items: center;
   gap: 0.6rem;
   padding: 0.5rem 0;
-
-  // Fade gauche/droite pour signaler clairement "ca defile".
-  &::before,
-  &::after {
-    content: '';
-    position: absolute;
-    top: 0;
-    bottom: 0;
-    width: 48px;
-    pointer-events: none;
-    z-index: 3;
-  }
-  &::before { left: 48px;  background: linear-gradient(to right, #0f0f1e 0%, transparent 100%); }
-  &::after  { right: 48px; background: linear-gradient(to left,  #0f0f1e 0%, transparent 100%); }
 }
 
 .carousel-track {

web/src/app/sidebar/sidebar.component.html

@@ -60,7 +60,7 @@
   </div>
 
   <div class="sidebar-footer">
-    <span class="version">Version 0.3.0</span>
+    <span class="version">Version 0.4.0</span>
   </div>
 
 </aside>