claude bugpass

ai/parser.py

@@ -61,7 +61,7 @@ def _call_llm_sync(system_prompt, user_prompt):
                 return extracted
         return None
     except Exception as e:
-        print(f"LLM error: {type(e).__name__}: {e}", flush=True)
+        print(f"LLM error ({AI_CONFIG['model']}): {type(e).__name__}: {e}", flush=True)
         return None
 
 
@@ -119,7 +119,7 @@ async def parse(user_input, interaction_type, retry_count=0, errors=None, histor
 
     response_text = await _call_llm(prompt_config["system"], user_prompt)
     if not response_text:
-        return {"error": "AI service unavailable", "user_input": user_input}
+        return {"error": f"AI service unavailable (model: {AI_CONFIG['model']})", "user_input": user_input}
 
     try:
         parsed = json.loads(response_text)

bot/commands/knowledge.py

@@ -17,10 +17,16 @@ from ai.parser import client
 # Configuration
 EPUBS_DIRECTORY = os.getenv("KNOWLEDGE_EMBEDDINGS_DIR", "./bot/data")
 TOP_K_CHUNKS = 5
-EMBEDDING_MODEL = "sentence-transformers/all-minilm-l12-v2"
 CHAT_MODEL = "deepseek/deepseek-v3.2"
 EMBEDDING_EXTENSION = ".embeddings.json"
 
+# Map embedding dimensions to the model that produced them
+EMBEDDING_MODELS_BY_DIM = {
+    384: "sentence-transformers/all-minilm-l12-v2",
+    4096: "qwen/qwen3-embedding-8b",
+}
+DEFAULT_EMBEDDING_MODEL = "sentence-transformers/all-minilm-l12-v2"
+
 # Cache for loaded embeddings: {file_path: (chunks, embeddings, metadata)}
 _knowledge_cache: Dict[str, Tuple[List[str], List[List[float]], dict]] = {}
 
@@ -79,9 +85,14 @@ def load_knowledge_base(
     return _knowledge_cache[file_path]
 
 
-def get_query_embedding(query: str) -> List[float]:
+def get_embedding_model_for_dim(dim: int) -> str:
+    """Get the correct embedding model for a given dimension."""
+    return EMBEDDING_MODELS_BY_DIM.get(dim, DEFAULT_EMBEDDING_MODEL)
+
+
+def get_query_embedding(query: str, model: str = DEFAULT_EMBEDDING_MODEL) -> List[float]:
     """Embed the user's question via OpenRouter."""
-    response = client.embeddings.create(model=EMBEDDING_MODEL, input=query)
+    response = client.embeddings.create(model=model, input=query)
     return response.data[0].embedding
 
 
@@ -271,8 +282,12 @@ async def handle_knowledge(message, session, parsed):
         await message.channel.send(f"🔍 Searching **{book_title}**...")
 
         try:
+            # Detect embedding dimension and use matching model
+            emb_dim = len(embeddings[0]) if embeddings else 384
+            embedding_model = get_embedding_model_for_dim(emb_dim)
+
             # Get query embedding and search
-            query_emb = get_query_embedding(query)
+            query_emb = get_query_embedding(query, model=embedding_model)
             relevant_chunks, scores = search_context(
                 query_emb, chunks, embeddings, TOP_K_CHUNKS
             )