Mentions and on-demand commands
Talk to the bot in PR comments — ask questions, teach it conventions, or trigger a fresh review.
Mention @revvu-ai anywhere in a PR comment and the bot reads what you wrote. There's no slash-command syntax to memorize and no fixed list of verbs. The bot decides what you meant from the sentence itself — a question, a convention to remember, or a request for a fresh review — and acts accordingly.
How mentions work
When you @-mention the bot in a PR comment, GitHub sends the comment to Revvu. The bot reacts with 👀 to confirm it received the mention, then reads the text and decides what to do. The reply lands in the same thread, just like a teammate's reply would.
- You write a comment containing @revvu-ai followed by what you want.
- The bot reacts with 👀 within a few seconds.
- The bot reads the comment and picks an action: answer the question, remember the rule, or queue a fresh review.
- The reply (or the new review) appears shortly after.
Three things you can do
The bot recognizes three intents from natural language. You don't have to phrase any of them a specific way — write the way you'd write to a teammate.
- Ask a question — Anything about the PR, the diff, or the surrounding code. The bot replies in the thread.
- Teach a learning — A convention, a rule, or a piece of context you want the bot to apply on future reviews of this repo. Stored in per-repo team memory.
- Trigger a re-review — Run the deep review again on the latest commit, even if you haven't pushed since the last review.
Examples
Every example below is something the bot understands as written. You can phrase them however feels natural.
- @revvu-ai how does the rate limiter handle bursts when the queue is empty?
- @revvu-ai we always use the AppError class for thrown errors, never raw Error
- @revvu-ai never use useEffect for data fetching — server components or React Query only
- @revvu-ai please review again
- @revvu-ai re-review the latest commit
- @revvu-ai what does this changeset actually break for existing users?
Learnings shape future reviews
When you teach the bot a convention, it stores the rule in per-repo team memory. The next review on that repo loads those learnings and feeds them into the analysis. Over a few weeks of use, the bot starts to feel like it knows your codebase — because it does.
Where the reply lands
The bot's reply goes into the same comment thread you mentioned it from. If you mentioned the bot at the top of the PR (an issue comment), the reply appears as another issue comment. If you mentioned it inline on a specific line of code (a review comment), the reply lands as a threaded reply to that comment. Use whichever surface fits the question — broad questions belong at the top of the PR, code-specific questions belong on the line.
What the bot won't do
The bot is a reviewer, not a code editor. It won't push commits, open PRs, or modify your branch. And it won't change its safety bar based on a learning — telling it 'don't flag SQL injection here' or 'stop warning about hardcoded secrets' won't work. Real security and correctness issues will still be raised regardless of what the team memory says.