AI pull request review that checks every finding.
HeyDeer reviews each pull request against your whole repository, verifies every finding against the code before it posts, and comments on the exact lines with a change you can commit. It has the highest precision on ReviewBench, GitHub’s open benchmark for AI code review: 99.0%.
No seat fees · 2-click install · Free plan
Everything a reviewer needs, on the lines it’s about
HeyDeer posts its findings as a GitHub pull request review, with an inline comment on each line a finding is about.
Severity and category
Every finding is Critical, High, Medium or Low, in Security, Correctness, Reliability, Performance or Maintainability. For example: “High (Reliability): Retry loop never stops on a 404.”
A change you can commit
Each finding explains what’s wrong and what it causes, then suggests an exact edit, with the reasoning and a diff folded underneath.
Breakage outside the diff
HeyDeer reads the change in the context of the whole repository, so it can flag code outside the diff that the change breaks.
A risk level for every PR
Each review rates the PR low, medium or high risk by how hard a missed mistake would be to contain, and says why.
A walkthrough for large PRs
Related changes are grouped and marked Must read, Should read or Skim. Changes that span the browser, server and database open with a diagram.
Honest about what it couldn’t check
Review notes say when something limited a review. A review with notes is incomplete, so HeyDeer never approves it.
Fewer comments that turn out to be wrong
A Standard review looks at the change from independent perspectives and verifies each finding against the code. Every finding carries a confidence level, Certain, Likely or Suspected, and your feedback scope decides which ones are posted.
Concise
For teams that read every comment: only findings HeyDeer is certain of, except Critical, which needs Likely. Low findings come from your repository checks only.
Balanced (default)
Critical findings at Suspected, High at Likely, and Medium and Low only when Certain.
Detailed
For teams whose coding agent verifies each finding before fixing it: Suspected for Critical and High, Likely for Medium and Low.
How a review runs
- 1
Install the GitHub App
Pick all repositories or only some. HeyDeer reads the whole repository from the first review.
- 2
Open or push to a PR
Review on every push, once per PR, or only when someone comments
@heydeer review. - 3
Read the review
Inline comments, a summary and the HeyDeer Review check land on the commit it reviewed.
- 4
Push the fixes
Follow-up reviews look only at what changed and recheck earlier findings. HeyDeer resolves its conversations once the code shows they’re fixed.
You decide when it reviews and what it blocks
Schedules and filters
Skip drafts, forks and dependency bots by default, or filter by label, author and team. Pull requests from coding agents are reviewed like any other.
Standard or Deep
Standard suits most pull requests. Deep spends longer on each step, for payments, authentication and other sensitive code. Deep needs a paid plan.
A check you can require
Turn on “Fail the check when issues are found” and require the HeyDeer Review check, and pull requests with open issues can’t merge.
Commands in any comment
@heydeer review,@heydeer full review,@heydeer pauseand@heydeer resumework in every connected repository.Automatic approval, if you want it
Off by default. Turn it on and HeyDeer approves complete reviews of low- and medium-risk PRs with no unresolved issues.
Insights
See findings per PR, findings by severity and what your team fixed or declined, by repository and author.
Frequently asked questions
Does HeyDeer work with GitLab or Bitbucket?
HeyDeer works with GitHub, for organizations and personal accounts. It installs as a GitHub App, and you can give it access to all repositories or only the ones you select.
Does HeyDeer store my code or train on it?
No. Every review works on a temporary clone that’s discarded when the work ends, and HeyDeer never uses your code, instructions or review results to train AI models.
How much does an AI pull request review cost?
HeyDeer charges credits for the work a review does, so there’s no fixed price per review, and Deep reviews use more than Standard. The Free plan includes 1,000 credits in the first month and 300 each month after; paid plans start at $9 a month. Failed reviews are refunded in full.
Will HeyDeer block my pull requests from merging?
Not by default. Findings don’t fail the HeyDeer Review check until you turn on “Fail the check when issues are found” and require the check in your branch rules.
Which AI models does HeyDeer use?
HeyDeer routes each part of a review, such as looking for findings, verifying them and applying repository checks, to the model that suits it. A single review can combine OpenAI, Anthropic, Kimi and GLM models.
What if I disagree with a finding?
Reply in the finding’s conversation with your reason, or resolve it. HeyDeer accepts the decision unless it has concrete evidence that it missed something serious, and an accepted decline isn’t posted again.
More from HeyDeer
- Fix with HeyDeerTurn findings into a tested fix pull request.
- AutopilotFixes findings, requests, failing checks and conflicts until the PR is ready.
- Rules and checksREVIEW.md, AGENTS.md and checks HeyDeer verifies on every PR.
- Coding agent PRsReview what Claude Code, Codex and Cursor write before it merges.
- Teams with reviewersHeyDeer checks the details so reviewers can focus on design.