Juan José Mata f491916411 Track failed LLM API calls in llm_usages table (#360) ...

* Track failed LLM API calls in llm_usages table

This commit adds comprehensive error tracking for failed LLM API calls:

- Updated LlmUsage model with helper methods to identify failed calls
  and retrieve error details (failed?, http_status_code, error_message)

- Modified Provider::Openai to record failed API calls with error metadata
  including HTTP status codes and error messages in both native and
  generic chat response methods

- Enhanced UsageRecorder concern with record_usage_error method to support
  error tracking for auto-categorization and auto-merchant detection

- Updated LLM usage UI to display failed calls with:
  - Red background highlighting for failed rows
  - Error indicator icon with "Failed" label
  - Interactive tooltip on hover showing error message and HTTP status code

Failed calls are now tracked with zero tokens and null cost, storing
error details in the metadata JSONB column for visibility and debugging.

* Dark mode fixes

---------

Co-authored-by: Claude <noreply@anthropic.com>

2025-11-22 02:15:20 +01:00

6.5 KiB

Raw Blame History

View Raw