Add LLM diagnostics: /api/test-llm endpoint + startup/error logging

- GET /api/test-llm: shows token status, model, api_base, and tests a real call
- Startup prints LLM config (token set/not-set, model, base_url) to logs
- LLM exceptions in execute-query now print to stdout (visible in container logs)

backend/api/demo.py

@@ -14,9 +14,13 @@ from __future__ import annotations
 
 import asyncio
 import json
+import logging
+import os
 import time
 from typing import AsyncIterator, Optional
 
+logger = logging.getLogger(__name__)
+
 from fastapi import APIRouter
 from pydantic import BaseModel
 from sse_starlette.sse import EventSourceResponse
@@ -49,6 +53,48 @@ from gepa.optimizer import get_gepa, QueryResult
 router = APIRouter()
 
 
+# ─── /api/test-llm ───────────────────────────────────────────────
+
+@router.get("/test-llm")
+async def test_llm():
+    """Diagnostic: test LLM connectivity and return result."""
+    from env.sql_env import _make_client, _MODEL
+    token = os.environ.get("HF_TOKEN", "")
+    api_base = os.environ.get("API_BASE_URL", "https://router.huggingface.co/v1")
+    token_preview = f"{token[:8]}..." if len(token) > 8 else ("(empty)" if not token else token)
+
+    try:
+        client = _make_client()
+        resp = await client.chat.completions.create(
+            model=_MODEL,
+            messages=[{"role": "user", "content": "Reply with just: OK"}],
+            temperature=0,
+            max_tokens=5,
+        )
+        result = resp.choices[0].message.content
+        return {
+            "ok": True,
+            "model": _MODEL,
+            "api_base": api_base,
+            "token_set": bool(token),
+            "token_preview": token_preview,
+            "response": result,
+        }
+    except Exception as e:
+        err = str(e)
+        if len(err) > 400 or '<html' in err.lower():
+            err = f"{type(e).__name__}: (response body too long, likely HTML error page)"
+        logger.error("test-llm failed: %s", err)
+        return {
+            "ok": False,
+            "model": _MODEL,
+            "api_base": api_base,
+            "token_set": bool(token),
+            "token_preview": token_preview,
+            "error": err,
+        }
+
+
 # ─── /api/init ────────────────────────────────────────────────────
 
 @router.get("/init")
@@ -172,6 +218,8 @@ async def execute_query_stream(req: ExecuteQueryRequest):
             except Exception as e:
                 # Format LLM exception concisely (avoid dumping full HTML 401 pages)
                 err_str = str(e)
+                logger.error("LLM call failed attempt=%d: %s: %s", attempt, type(e).__name__, err_str[:200])
+                print(f"[execute-query] LLM error attempt={attempt}: {type(e).__name__}: {err_str[:200]}", flush=True)
                 if len(err_str) > 300 or '<html' in err_str.lower():
                     err_str = f"LLM API error: {type(e).__name__} (check HF_TOKEN / model availability)"
                 yield {"data": json.dumps({"type": "error", "message": err_str, "error_class": "other"})}

backend/main.py

@@ -13,9 +13,12 @@ Environment variables:
 
 from __future__ import annotations
 
+import logging
 import os
 from pathlib import Path
 
+logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(name)s: %(message)s")
+
 from fastapi import FastAPI
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.staticfiles import StaticFiles
@@ -83,6 +86,13 @@ async def startup_event():
     except Exception as e:
         print(f"Warning: database seed failed: {e}")
 
+    # Log LLM config so it's visible in container logs
+    token = os.environ.get("HF_TOKEN", "")
+    api_base = os.environ.get("API_BASE_URL", "https://router.huggingface.co/v1")
+    model = os.environ.get("MODEL_NAME", "Qwen/Qwen2.5-72B-Instruct")
+    token_status = f"set ({len(token)} chars)" if token else "NOT SET"
+    print(f"[startup] LLM config: base_url={api_base} model={model} HF_TOKEN={token_status}", flush=True)
+
 
 # ─── Static files (frontend) — mount last ─────────────────────────