testing journey agent creation

app/api/routes/agent_setup.py

@@ -31,7 +31,7 @@ from typing import Any
 
 from langchain_core.messages import AIMessage, HumanMessage, SystemMessage, ToolMessage
 
-from app.agents.filesystem_agent import FILESYSTEM_TOOLS
+from app.agents.filesystem_agent import make_directory_tools
 from app.config.settings import settings
 from app.core.langfuse_client import compile_prompt, extract_usage, get_langfuse, get_prompt_or_fallback
 from app.core.llm import get_llm
@@ -273,7 +273,7 @@ async def _call_llm_with_tools(
     _span = _span_ctx.__enter__() if _span_ctx else None
 
     try:
-        for _ in range(_MAX_TOOL_STEPS):
+        for step in range(_MAX_TOOL_STEPS):
             _gen_ctx = (
                 lf.start_as_current_observation(
                     as_type="generation",
@@ -290,12 +290,24 @@ async def _call_llm_with_tools(
                 _gen.update(output=_as_text(response.content), usage=extract_usage(response))
                 _gen_ctx.__exit__(None, None, None)
 
+            resp_text = _as_text(response.content)
+
+            # Guard against empty responses (e.g. model returned finish_reason
+            # 'error' which LiteLLM maps to 'stop' with empty content).
+            if not response.tool_calls and not resp_text.strip():
+                logger.warning(
+                    "agent_setup: journey LLM returned empty response at step %d — retrying",
+                    step,
+                )
+                # Drop the empty AIMessage so we don't pollute history, and retry.
+                continue
+
             messages.append(response)
 
             if not response.tool_calls:
                 if _span:
-                    _span.update(output=_as_text(response.content))
-                return _as_text(response.content)
+                    _span.update(output=resp_text)
+                return resp_text
 
             for call in response.tool_calls:
                 call_name = str(call.get("name", ""))
@@ -324,7 +336,10 @@ async def _call_llm_with_tools(
         final_text = _as_text(final.content)
         if _span:
             _span.update(output=final_text)
-        return final_text
+        return final_text or (
+            "Sorry, I had trouble processing the files. "
+            "Could you try again? If the issue persists, the files might be too large for me to analyse."
+        )
     finally:
         if _span_ctx:
             _span_ctx.__exit__(None, None, None)
@@ -372,7 +387,7 @@ async def handle_journey_start(
     ai_reply = await _call_llm_with_tools(
         system_prompt=system_prompt,
         history=seed_history,
-        tools=list(FILESYSTEM_TOOLS),
+        tools=make_directory_tools(directory),
         user_id=user_id,
         session_id=session_id,
         langfuse_prompt=langfuse_prompt,
@@ -436,10 +451,11 @@ async def handle_journey_message(
     session.history.append({"role": "user", "content": message})
 
     # Call the LLM with tools.
+    session_tools = make_directory_tools(session.directory)
     ai_reply = await _call_llm_with_tools(
         system_prompt=session.system_prompt,
         history=session.history,
-        tools=list(FILESYSTEM_TOOLS),
+        tools=session_tools,
         user_id=session.user_id,
         session_id=session_id,
         langfuse_prompt=session.langfuse_prompt,
@@ -464,7 +480,7 @@ async def handle_journey_message(
             nudge_reply = await _call_llm_with_tools(
                 system_prompt=session.system_prompt,
                 history=session.history,
-                tools=list(FILESYSTEM_TOOLS),
+                tools=session_tools,
                 user_id=session.user_id,
                 session_id=session_id,
                 langfuse_prompt=session.langfuse_prompt,