fix: Add max_tokens=16384 to LLM requests (OpenWebUI defaults to 1000, truncating pipeline JSON)

.env.example

@@ -29,6 +29,9 @@ LLM_FALLBACK_MODEL=fyn-qwen35-chat
 #LLM_STAGE5_MODEL=fyn-qwen35-thinking
 #LLM_STAGE5_MODALITY=thinking
 
+# Max tokens for LLM responses (OpenWebUI defaults to 1000 — pipeline needs much more)
+LLM_MAX_TOKENS=16384
+
 # Embedding endpoint (Ollama container in the compose stack)
 EMBEDDING_API_URL=http://chrysopedia-ollama:11434/v1
 EMBEDDING_MODEL=nomic-embed-text

backend/config.py

@@ -43,6 +43,9 @@ class Settings(BaseSettings):
     llm_stage5_model: str | None = None  # synthesis — thinking model recommended
     llm_stage5_modality: str = "chat"
 
+    # Max tokens for LLM responses (OpenWebUI defaults to 1000 which truncates pipeline JSON)
+    llm_max_tokens: int = 16384
+
     # Embedding endpoint
     embedding_api_url: str = "http://localhost:11434/v1"
     embedding_model: str = "nomic-embed-text"

backend/pipeline/llm_client.py

@@ -136,6 +136,7 @@ class LLMClient:
             response = self._primary.chat.completions.create(
                 model=primary_model,
                 messages=messages,
+                max_tokens=self.settings.llm_max_tokens,
                 **kwargs,
             )
             raw = response.choices[0].message.content or ""
@@ -156,6 +157,7 @@ class LLMClient:
             response = self._fallback.chat.completions.create(
                 model=fallback_model,
                 messages=messages,
+                max_tokens=self.settings.llm_max_tokens,
                 **kwargs,
             )
             raw = response.choices[0].message.content or ""