Aido 1.2.1 improves reliability by retrying transient AI provider errors

Aido, the open-source multi-provider AI assistant for GitHub code review, has released version 1.2.1. The patch release focuses on reliability, with fixes surfaced by live use on Aido's first external adopter.

Aido lets development teams review, summarize, explain, and document code by commenting on a pull request. It works with Gemini, ChatGPT, and Claude, and runs inside GitHub Actions on a team's own provider keys.

Retrying transient provider errors

AI provider calls now retry temporary failures with exponential backoff, covering the common HTTP status codes for rate limiting and service unavailability. Previously a single transient error, such as the 503 responses that Gemini's free tier can return under load, would abort an entire command. Retrying these blips makes every Aido command more dependable.

Reviews degrade gracefully

The review command makes two provider calls: one for the main review and a second for inline suggestions. Before this release, a temporary failure on the second call discarded the whole review, including the part that had already succeeded. Now the review still posts, with a note that inline suggestions were skipped and can be retried.

Aido's earlier "fail loudly" behavior is preserved: exhausted retries and genuine, non-transient errors still stop with a clear, actionable message rather than failing silently.

Version 1.2.1 is available now on GitHub. Remote installs upgrade by bumping the version tag in their workflow.