Enhanced LLM error diagnostics | Generated by AI

Home 2026.05

All good. Here’s a summary of what was changed and why:

Changes made

ww/llm/openrouter_client.py — LLM client layer

Problem: When call_openrouter_api_with_messages got an empty content response (HTTP 200 but message.content is None or ""), it just returned the falsy value silently. When it got a non-200 response, the error message only showed the status code and raw response body — no context about model, proxy, or tokens.

Fix: Added two improvements:

  1. _check_proxy() — new helper that checks proxy env vars and tests TCP connectivity to the proxy host:port. Returns a human-readable string like "HTTP_PROXY=http://127.0.0.1:7890 (port 7890 reachable)" or "UNREACHABLE: ...".

  2. Empty content detection — after a successful HTTP 200, parses the JSON body and checks if content is falsy. If so, raises an exception with: model name, max_tokens, finish_reason, HTTP status, proxy status, and full response body (first 1000 chars). This covers the case that hit you — when the model returns content: null.

  3. Enhanced HTTP error — now includes model, max_tokens, proxy status alongside the status code and response body.

ww/note/create_note_utils.py — caller layer

Problem: _call_llm_or_exit swallowed all detail. The old code:

result = call_openrouter_api(prompt, max_tokens=max_tokens)
if not result:
    print("Failed to generate title with max 6 words. Exit.")
    sys.exit(1)

No model name, no prompt preview, no exception details.

Fix: Wrapped in a try/except that prints:

And for the empty-result path (which shouldn’t happen anymore since the LLM client now raises on empty content):

Now if ww note fails on title generation again, you’ll see the model name, the prompt it sent, the HTTP status, the finish_reason from the API, and your proxy connectivity status — everything needed to diagnose what went wrong.

Back Donate