Skip to main content

Comment Answers

Comment Answers are designed for follow-up discussion inside merge-request threads. This capability is different from Code Review: instead of analyzing the entire diff, it replies to a specific note or discussion.

When Buzzbin answers​

In the current product behavior, answers are mention-gated. That means Buzzbin only acts when the bot is explicitly called inside an MR note.

In practice:

  • it works on Merge Request notes
  • it usually answers only when @buzzbin appears in the note body
  • it ignores system notes
  • it does not answer notes authored by the bot itself

When you should not expect an answer​

These cases usually produce no answer:

  • you replied in the thread but did not mention the bot
  • the Comment Answers capability is disabled for the repository
  • the note belongs to an Issue or another object instead of an MR
  • the note is a system note
  • the wallet has no available credit or the run is blocked by guardrails

Key Comment Answer settings​

These Review Config fields affect answer behavior:

  • whether the capability is enabled at all
  • the Answer model
  • the answer language
  • the Answer only on mentions toggle

By default, Answer only on mentions stays enabled so Buzzbin does not jump into every thread and answer costs remain more predictable.

Answer language and model​

  • answers can be in Persian or English
  • the answer language usually inherits from Code Review unless overridden
  • Buzzbin uses the answer model for replies, not the review model

That split matters because a comment answer is usually lighter and more latency-sensitive than a full diff review.

Behavior examples​

Answer is generated​

@buzzbin Is this null check enough, or do we still have a crash path here?

Answer is not generated​

I think this section still needs another look.

If you want Buzzbin to join the second discussion, mention the bot explicitly.

Queue and cost impact​

Comment Answers create their own Review Jobs and appear in run history. That means they:

  • have their own cost
  • can be blocked by credit limits or guardrails
  • can be inspected through Review Jobs and Webhook Deliveries

Next step​

If you want to see how findings and review output are structured, continue to Findings, severity, and expected output.