Fine-tune pull requests inside GitHub.

Your team already reviews code in GitHub. ProPR keeps refinement there too: it opens PRs automatically and turns review comments into follow-up commits on the same branch.

Comment-to-commit Same PR branch

PR Comment "fix the padding"

Isolated RunAgent applies fix

New CommitAdded to branch

Review continues in GitHub

A review comment starts the isolated run, and the resulting commit lands on the same PR branch.

Line-level context

Line-level review comments automatically include their specific file and line context, so the agent pinpoints exactly what needs to change without you explaining it.

Image-aware comments

Paste screenshots directly into your PR comments. ProPR makes them available to the agent for visual bug fixes and UI polish.

Intentional triggers

Choose exactly which comments or prefixes trigger an agent run. Casual team discussions in the PR stay separate from expensive LLM tasks.

Single-branch history

ProPR pushes follow-up commits directly to the existing PR branch, preserving review history.

A recent human follow-up request on public ProPR pull request 1613. — Comment to commit on public propr #1613.

ProPR replying with the follow-up commit and passing verification. — Comment to commit on public propr #1613.

The PR can ask for more than a small fix.

When review finds a problem, continue on the same pull request. Handle follow-up fixes, branch updates, merge conflicts, stronger review-fix loops, and rollbacks entirely through PR commands.

Apply review findings

Human comments become direct follow-up work, and /fix applies the unprocessed AI review comments left by /review.

Ultrafix loop

/ultrafix alternates review and fix cycles until it hits a target score, a cycle limit, or a human stop signal.

Branch updates

Update a stale PR branch from the review thread; ProPR routes conflicts through safe cleanup when needed.

Revert flow

Completion comments can include undo links that back out a follow-up commit.

Choose the right model for the job.

Choose separate agents and models for planning, implementation, and refinement from one self-hosted workflow.

See supported agents →