modified: blueprints/chat.py

modified:   services/llm_service.py
	modified:   services/rag_service.py
	modified:   static/js/chat.js

blueprints/chat.py

@@ -47,6 +47,22 @@ def get_messages(session_id):
     return jsonify([m.to_dict() for m in session.messages])
 
 
+# ── Diagnostics ─────────────────────────────────────────────────────────────
+
+@chat_bp.route("/ping-llm", methods=["GET"])
+@login_required
+def ping_llm():
+    """Test LM Studio connectivity. Returns config + a short completion."""
+    url = current_app.config.get("LM_STUDIO_URL", "")
+    model = current_app.config.get("LM_STUDIO_MODEL", "")
+    embed_model = current_app.config.get("LM_STUDIO_EMBEDDING_MODEL", "")
+    try:
+        reply = llm_service.ask(user_message="Reply with exactly: OK", context_chunks=[], history=[])
+        return jsonify({"status": "ok", "reply": reply, "url": url, "model": model, "embed_model": embed_model})
+    except Exception as e:
+        return jsonify({"status": "error", "error": str(e), "url": url, "model": model, "embed_model": embed_model}), 502
+
+
 # ── Main chat ────────────────────────────────────────────────────────────────
 
 @chat_bp.route("/sessions/<int:session_id>/ask", methods=["POST"])
@@ -64,31 +80,33 @@ def ask(session_id):
     doc_ids = [r["id"] for r in context_refs if r.get("type") == "doc"]
     url_ids = [r["id"] for r in context_refs if r.get("type") == "url"]
 
-    # RAG lookup
+    # RAG lookup — failures are non-fatal (chat continues without context)
     chunks = []
-    if doc_ids:
+    try:
-        chunks += rag_service.similarity_search(
+        if doc_ids:
-            query=message,
+            chunks += rag_service.similarity_search(
-            user_id=current_user.id,
+                query=message,
-            source_ids=doc_ids,
+                user_id=current_user.id,
-            source_type="doc",
+                source_ids=doc_ids,
-            top_k=current_app.config["RAG_TOP_K"],
+                source_type="doc",
-        )
+                top_k=current_app.config["RAG_TOP_K"],
-    if url_ids:
+            )
-        chunks += rag_service.similarity_search(
+        if url_ids:
-            query=message,
+            chunks += rag_service.similarity_search(
-            user_id=current_user.id,
+                query=message,
-            source_ids=url_ids,
+                user_id=current_user.id,
-            source_type="url",
+                source_ids=url_ids,
-            top_k=current_app.config["RAG_TOP_K"],
+                source_type="url",
-        )
+                top_k=current_app.config["RAG_TOP_K"],
-    # If no specific ids given, search all user context
+            )
-    if not context_refs:
+        if not context_refs:
-        chunks = rag_service.similarity_search(
+            chunks = rag_service.similarity_search(
-            query=message,
+                query=message,
-            user_id=current_user.id,
+                user_id=current_user.id,
-            top_k=current_app.config["RAG_TOP_K"],
+                top_k=current_app.config["RAG_TOP_K"],
-        )
+            )
+    except Exception as e:
+        current_app.logger.warning(f"RAG lookup failed, continuing without context: {e}")
 
     # Build history (last 10 messages for context window)
     history = [

services/llm_service.py

@@ -21,13 +21,14 @@ def _get_client():
 
     if provider == "openai":
         api_key = current_app.config.get("OPENAI_API_KEY", "")
-        _client = openai.OpenAI(api_key=api_key)
+        _client = openai.OpenAI(api_key=api_key, timeout=60.0)
     else:
         # LM Studio or any OpenAI-compatible endpoint
         base_url = current_app.config.get("LM_STUDIO_URL", "http://localhost:1234")
         _client = openai.OpenAI(
             base_url=f"{base_url.rstrip('/')}/v1",
             api_key="lm-studio",
+            timeout=60.0,
         )
 
     return _client

services/rag_service.py

@@ -20,7 +20,11 @@ class LMStudioEmbeddingFunction(EmbeddingFunction):
     """ChromaDB-compatible embedding function that calls LM Studio's /v1/embeddings."""
 
     def __init__(self, base_url: str, model: str):
-        self._client = OpenAI(base_url=f"{base_url}/v1", api_key="lm-studio")
+        self._client = OpenAI(
+            base_url=f"{base_url}/v1",
+            api_key="lm-studio",
+            timeout=60.0,
+        )
         self._model = model
 
     def __call__(self, input: Documents) -> Embeddings:

static/js/chat.js

@@ -28,20 +28,32 @@ export class Chat {
     // Typing indicator
     const typingId = this._appendTyping();
 
-    const res = await fetch(`/api/chat/sessions/${sessionId}/ask`, {
+    try {
-      method: 'POST',
+      const res = await fetch(`/api/chat/sessions/${sessionId}/ask`, {
-      headers: { 'Content-Type': 'application/json' },
+        method: 'POST',
-      body: JSON.stringify({ message, context_ids: contextIds }),
+        headers: { 'Content-Type': 'application/json' },
-    });
+        body: JSON.stringify({ message, context_ids: contextIds }),
+      });
 
-    this._removeTyping(typingId);
+      this._removeTyping(typingId);
 
-    const data = await res.json();
+      let data;
+      try {
+        data = await res.json();
+      } catch {
+        this._appendBubble('assistant', 'Server error — could not parse response.', true);
+        this._scrollBottom();
+        return;
+      }
 
-    if (!res.ok) {
+      if (!res.ok) {
-      this._appendBubble('assistant', `Error: ${data.error || 'Unknown error'}`, true);
+        this._appendBubble('assistant', `Error: ${data.error || 'Unknown error'}`, true);
-    } else {
+      } else {
-      this._appendBubble('assistant', data.reply);
+        this._appendBubble('assistant', data.reply);
+      }
+    } catch (err) {
+      this._removeTyping(typingId);
+      this._appendBubble('assistant', `Network error: ${err.message}`, true);
     }
 
     this._scrollBottom();