Review the change.
Understand the codebase.

OpenReview reviews pull requests using your whole repository, not only the diff. Every finding quotes the code it depends on, and nothing is posted until it has been verified.

The OpenReview pull request page for acme/shop #42: the reviewed diff with a P0 finding shown inline at the changed line, a "Needs changes" verdict and a confidence score of 2 out of 5.

Reviews pull requests on

  • GitHub
  • GitHub Enterprise
  • GitLab
  • Azure DevOps
  • Bitbucket
  • Gitea
  • Perforce
Evidence

Evidence on every finding

Each issue cites the exact lines it depends on, pinned to the commit, with a committable suggestion when the fix is small. A finding that cannot quote the reviewed revision is dropped.

Learn how reviews work →
A verified P0 finding with the file and line it depends on, its impact, three quoted lines of evidence and a suggested fix.
Summary

One summary, kept current

A single summary comment with an overview, changed files, issues, a confidence score and a sequence diagram is updated as the pull request moves. New pushes are reviewed against what changed.

See what a review posts →
The review summary: what changed, affected areas and a recommendation, followed by a history of reviews for each pushed revision.
Chat

Ask your codebase

Chat, the API, MCP and Slack all answer questions with citations that must match the indexed source.

Learn about codebase chat →
Codebase chat: a question about where checkout totals are computed, answered with two citations that quote the indexed source.
Fixes

Fix with your agent

Hand findings to Codex, Claude Code or Cursor Agent on your machine, or ask @openreview for a validated fix.

Set up the agent bridge →
A finding on the Findings page with controls to send it to Codex, Claude Code or Cursor Agent, copy a fix prompt, or open an agent with the prompt pre-filled.
Analytics

See what reviews are worth

Track reviews, verified findings, addressed rate, time saved and model cost by repository, team and author, and export it as CSV.

Read the quickstart →
The analytics page: pull requests, reviews, verified findings and model cost for the period, daily review activity, findings by severity and a trend of reviews and findings per day.

Built for reviews you can trust

Incremental re-reviews

New pushes are reviewed against what changed, and threads for fixed findings resolve themselves.

Configuration as code

Rules and settings live in .openreview.json, read from the base branch so a PR cannot weaken its own review.

Many providers

GitHub, GitHub Enterprise, GitLab, Azure DevOps, Bitbucket Cloud and Data Center, Gitea and Perforce.

Your model, your budget

Bring Anthropic, OpenAI or a compatible endpoint. A hard inference budget and per-task usage accounting keep spend predictable.

How it works

  1. 01

    Index the repository

    OpenReview parses your default branch into files, symbols and a cross-file call and import graph, with optional semantic embeddings.

  2. 02

    Investigate each change

    The diff is read alongside its callers, tests, past findings, your rules and any linked requirements, by specialists or a tool-using agent.

  3. 03

    Verify before posting

    A finding needs exact quotes from the reviewed revision and survives an independent check. Unsupported claims are dropped.

  4. 04

    Learn from your team

    Reactions, feedback and recurring human review comments become memory that your admins can inspect and control.

Run it on your infrastructure

Your Next.js host, your Convex backend and your model provider. Code goes only to the model you configure, and secret-looking values are redacted first.

Read the self-hosting guide →
Terminal
$ pnpm install
$ CONVEX_AGENT_MODE=anonymous pnpm exec convex dev --once
$ pnpm setup:local
$ pnpm backend
$ pnpm dev

Self-hosted or cloud

AspectSelf-hostedCloud
Where it runsYour Next.js host and Convex backendA hosted OpenReview deployment
Model and keysYours, with your own budgetProvided by the operator
Review limitsNonePlan quota per seat
SetupSelf-hosting guideSign in and install the app

Questions

Does OpenReview run our code?

No. Reviews read source and never execute it. Test results appear only when a paired runtime worker records them for that commit.

Where does our code go?

Self-hosted, it stays in your Convex deployment and goes only to the model provider you configure. Secret-looking values are redacted before model calls.

Can a pull request turn its own review off?

No. Repository configuration is read from the base branch, and admins can lock settings.

Will it flood our PRs with comments?

Findings must pass confidence, severity and evidence thresholds. You choose the maximum comment count, strictness and categories, and learned preferences reduce repeat noise.

Which models are supported?

Claude and OpenAI models, or any OpenAI-compatible endpoint, with per-task model routing.

Try OpenReview now.

OpenReview: repository-aware pull request reviews