How it works

A pull request moves from discovery to a review to learning signals that shape the next review. Here is the whole loop.

Step 1
Watch a repo
Register a GitHub or Azure DevOps repository as a watched repo in Settings. PR Bot discovers its open pull requests.
Step 2
Review the PR
A Temporal workflow clones the repo, fetches existing comments and runs the AI review through your MCP provider.
Step 3
Post the result
The verdict, Issues and PR-level notes are stored and posted back to the PR as bot threads, unless it is a dry run.
Step 4
Watch the replies
Review monitoring polls the PR for new human replies to bot threads, new commits and state changes.
Step 5
Approve what was learned
Each reply becomes a classified learning signal. Written replies wait in the approval queue for a person to approve or reject.
Step 6
Review better next time
Approved signals build the repo context, which goes into every later review of that repo.

Your repos, your servers, your AI back end

Run PR Bot with Docker Compose or as a macOS desktop app. Reviews go through the MCP provider you configure.