Skip to content
Community editionSelf-hosted review pathDear investorsMemo & roadmap

Merge
with
confidence.

Critique is automated code verification for GitHub pull requests: we spin up an ephemeral sandbox, run your code, catch the breakage that kills builds, and give you sandbox-powered safe merging without seat-based pricing. We are also opening a separate self-hosted community edition around the core review workflow. The Coding Agent API ships the same OpenCode runtime over HTTP for CI, internal agents, and fix workflows.

Install on GitHubCoding Agent APIPricing
Community edition details
Sandbox verification before mergeBuild, type, and policy proofNo seat-based review taxCommunity edition live on GitHub
ReviewRemedyChat

Review status

Merge blocked.

Scout · 4 specialists · lead verdict

This PR weakens tenant scoping, risks double-applying billing reconciliation, and slows the context builder in the hot path. Merge is not recommended until those findings are resolved.

github-nativespecialist handoffcheck run output
sandbox verification·safe merging·risk score·build proof·merge policy as code·typecheck proof·verified repair·github pr verification·vercel-safe merges·github-native checks·sandbox verification·safe merging·risk score·build proof·merge policy as code·typecheck proof·verified repair·github pr verification·vercel-safe merges·github-native checks·sandbox verification·safe merging·risk score·build proof·merge policy as code·typecheck proof·verified repair·github pr verification·vercel-safe merges·github-native checks·

Community edition

Open source, built around the review core.

Critique now has a separate public community repo for teams that want to self-host the main GitHub PR review workflow. The boundary is deliberate: keep the merge-facing product open, keep private hosted systems out, and make the public repo operable on its own terms.

Open community pageGitHub repo

Self-host the main review loop.

The community edition is being shaped around the GitHub pull request review path teams actually need to operate themselves.

Keep hosted-only systems out of the boundary.

Private inference billing and company-only cloud internals stay outside the public repo so self-hosting stays legible.

Open in controlled slices.

Routes, review primitives, and operator setup move over deliberately instead of dumping a private monolith into public history.

Product map

One platform, every surface exposed.

Critique is more than the GitHub check. Use Intake before a bug reaches engineering, Review runs and Change Passports at the merge boundary, then the Platform, Merge Gate, Coding Agent, and Inference APIs when your own systems need the same controls.

Bug intake

Intake

Turn vague user reports into investigated tickets, triage, and agent-ready handoffs for the next fix path.

GET /intake/widget.js
Agent execution

Coding Agent API

Queue OpenCode sandbox runs over HTTP, preview credits, stream events, and open draft PRs from your tools.

POST /api/v1/coding-agent/runs
Model access

Inference API

OpenAI-compatible chat completions on Critique credits, with private rates or 75% off log retention pricing.

POST /api/v1/chat/completions
Merge control

Merge Gate API

Read review state, enforce policy, and wire signed lifecycle events into CI or release automation.

GET /api/v1/openapi
Review system

Review runs

Sandbox-backed PR review, specialist lanes, risk scoring, and one GitHub-native verdict per change.

POST /api/v1/reviews
Evidence layer

Change Passports

Audit provenance, findings, gate decisions, repair proof, and incident memory for each pull request.

GET /api/v1/passports

Coding Agent API

Ship repo work over HTTP — same runtime as Builder.

Start agent runs from CI, cron, or internal tools. Critique clones the repository, runs OpenCode in a sandbox, returns patch and timeline events, and can open a draft PR you hand to merge-grade review.

Explore the APIAPI referenceCompare vs Devin

Preview credits before you queue

POST with "preview": true returns model floor, balance, and blocked state — no sandbox spawned.

Same job as Builder

Workspace UI and HTTP API read and write the same builderJob, with links back to stream, status, and passport.

Warm sessions + SSE

Follow-ups stay in one OpenCode session; live activity streams over GET …/stream while a turn runs.

Draft PR → review

Publish a branch, open the PR, and jump to Change Passport review without re-entering context.

POST /api/v1/coding-agent/runs

{ "repository": "acme/web", "prompt": "…", "preview": true, "billing": { "mode": "managed" } }

Skills marketplace for agent reviewers.

Browse installable review skills, publish your own lenses, and route specialized judgment into every PR.

Open marketplacePublish a skill

Find review lenses for the codebase in front of you.

Security, platform migrations, billing, agent systems, performance, and team-specific policies install as versioned SKILL.md packages.

Enter

Publish the judgment your best reviewer keeps repeating.

Turn local review heuristics into a portable skill with tags, changelog, install commands, and account-gated releases.

Enter

Rank skills by merge impact.

See installs, acceptance, and skill performance so the marketplace rewards useful lenses instead of loud descriptions.

Enter

Start with the built-in critique-review skill.

The official review lens adds intake, triage, stack-aware specialists, and stricter finding contracts.

Enter

Install the same lens across every agent runtime.

Portable commands and bundles keep Codex, Cursor, Claude Code, and Critique Chat speaking the same review language.

Enter

Install

Drop a versioned review lens into Codex, Cursor, Claude Code, or Critique Chat.

Tune

Pin team-approved versions and keep experimental skills off production reviews.

Publish

Package your strongest reviewer instincts for the next repository.

Latest updates

Agent-owner tools are live.

Start with the Coding Agent API and Platform REST (`POST /api/v1/reviews`), then skills, BYOA, BYOK, and the v6 ship log — everything on this page links to the surfaces we ship in production.

Community

Community edition is live.

The self-hosted open-source boundary now has its own public repo, roadmap, and announcement page.

Open update
Agent API

Coding Agent API is live.

HTTP runs on OpenCode + E2B: idempotent creates, credit preview, signed webhooks, cancel/status endpoints, and OpenRouter BYOK.

Open update
v6.0.0

Platform API and calmer chrome.

Queue reviews over REST, enforce MCP scopes, subscribe to signed lifecycle webhooks, and open Critique in light mode by default — read the v6 outcome bullets on the ship log.

Open update
v5.1.0

Critique built Critique.

Autonomous agents shipped v5.1 in one window — transparent review runs, one Coding Agent product, credit preview, and repo-first Platform. Read the ship essay.

Open update
Skill upgrade

critique-review now reviews with depth.

The free built-in skill now adds intake and triage, stack lenses, and a stricter findings contract for serious PR review.

Open update
BYOA

Bring your own agent.

Queue Cursor, Claude, or Codex-style runs from completed reviews while vendor keys stay under your account.

Open update
BYOK

Bring your own key.

Use OpenRouter or CrofAI direct billing while Critique keeps orchestration, sandboxes, and ledgers in one place.

Open update
Ship log

Read the full changelog.

Marketplace, PR dashboard, skills, connections, and every production-facing update in one page.

Open update

The product is automated code verification. Review is one layer.

The merge boundary is the product. Critique records provenance and risk, proves a block with evidence, enforces merge policy, verifies repairs in a sandbox, and feeds incident learnings back into the rules. Review, chat, and Builder are layers inside that loop.

Every PR gets a verification record.

How should this work?
Keep chat light. Make Builder show receipts.
Switch to Builder

One PR-level record holds provenance, risk, gate events, evidence runs, the merge decision, repair proof, and memory. Internally we call it the Change Passport.

A block has to cite evidence.

components/home/page.tsx

+ product panel stack

+ unique abstract surfaces

- generic review slop

The evidence contract links every blocking decision to the finding behind it. A verdict is a claim you can audit and replay, not an opinion in a thread.

Merge policy is code.

tests84%

Dashboard or repo-file policy runs in dry-run, warn, or enforce. Operator overrides record who allowed the merge and why, then patch the GitHub check.

Repairs ship with proof.

Chat

free

Builder

38

Remedy stores a proof bundle: patch hash, validation, and a verification linkage. A fix is done because the proof says so.

Incidents teach the gate what to catch next time.

feat/home-redesign
diff ready
ship branch

Findings memory and incident learnings from Sentry, Linear, Jira, and Vercel feed back into the rules. The boundary gets smarter with every postmortem.

Scout the diff.
Split the risk.
Publish the verdict.

The review loop is the product loop: frame the change, verify risk with specialists, then post one grounded output back to GitHub.

Scout frames the review problem.

It reads the diff, nearby files, and test surface first so the review starts with context instead of generic commentary.

Workspace statediff first
Search repo history
Explain affected files
Keep thread free

Specialists verify the risk.

Security, billing, architecture, and performance lanes open only when the pull request actually warrants them.

Workspace stateparallel lanes
Builder lane selected
Repository context retained
Credits preview visible

GitHub gets one readable verdict.

The lead reviewer writes the final output as a technical review artifact your team can merge or block against.

Workspace statepublish verdict
Changed files listed
Diff summary attached
Validation status readable

Verify your code, not just review it.

Connect GitHub and the verification queue fills in. Run merge policy in dry-run first, then enforce when you are ready. No check renames, no forced gating, no rewrite of your existing automation.

Install on GitHubSee the platform
Verification record per PRProvenance plus risk scoreMerge policy as codeVerified repair with proofFindings and incident memoryGitHub-native check runs

Keep exploring

Full sitemap ↗

Everything you need to evaluate AI review and automated code verification.

Critique v4

Automated code verification at the merge boundary.

Sandbox-powered safe merging, Control Board policy, verification records per PR, and why comment-only review bots are not enough for agent-authored code.

Open

PR control

PR control for teams buried in pull requests.

Gate slop before review spend, operate from one Control Board, and keep an auditable verification record per PR.

Open

Git control

Git control without replacing Git.

Govern what merges on GitHub — agent PRs, policy, and proof — for platform and release engineering.

Open

All guides

Guides hub for operators.

PR control, git control, automated code verification, and AI review — one index.

Open

About

Not just a code review CI tool.

The real verification layer for pull requests: sandbox runs, policy, memory, delivery, and proof on one PR record.

Open

Definitive guide

AI code review, without the hype.

What AI code review is, how multi-agent review actually works, how to evaluate tools on your own PRs, and a 6-step rollout playbook.

Open

Tool comparison

Best AI code review tools in 2026.

Nine tools ranked, a full feature matrix, honest callouts of where each competitor wins, and direct vs-Critique deep dives.

Open

Pricing + credits

Transparent per-PR pricing.

Solo $19/mo, Pro $49/mo, Team $149/mo — shared across the team. Student and OSS maintainers start at $5/mo student/OSS with unlimited indexing and no CodeRabbit-style seat tax.

Open

Free calculator

What does PR review cost?

Estimate monthly reviewer hours, engineering cost, and the leverage of putting an AI review layer on one busy GitHub repo first.

Open

Head-to-head comparisons

See the ranked list

v6.0.0

Critique v6 — REST reviews, lifecycle webhooks, lighter UI.

Queue PR reviews over HTTP, enforce MCP scopes on ten tools, register signed installation webhooks, browse OpenAPI — and land in light mode by default instead of dark green marketing chrome.

Open

v5.1.0

Critique v5.1 — autonomous ship, operator-first review.

Transparent OpenCode controllers, one Coding Agent across Builder and API, credit preview, marketplace attribution, and repo-first Platform — mostly agent-built.

Open

Open source

Hundreds of PRs, no time to review?

PR control for foundations and ecosystems with hundreds of PRs — Pro/Team for volume, OSS lane for verified maintainers.

Open

PR operations

Manage pull requests at scale.

Weekly operating rhythm, verification queue triage, and PR management when agents flood your repos.

Open

Pricing research

AI code review pricing in 2026.

CodeRabbit seats, Cursor Bugbot usage billing, GitHub Copilot Actions-minute costs, and Critique shared credits compared in one buyer guide.

Open

Build failures

Vercel build failed on pull request?

Catch TypeScript compile errors, circular imports, bad dynamic imports, and broken Next.js builds before merge with local build checks and ephemeral sandbox verification.

Open

Security checklist

Review AI-generated code safely.

A practical PR checklist for auth, data access, secrets, dependencies, prompt injection, and safe remediation before merge.

Open

Trending now

AI code review trends in 2026.

What GitHub, Anthropic, Cursor, and OpenAI are signalling right now: agentic review, repo context, usage-based pricing, and verification-first buying criteria.

Open

Cursor BYOA

Composer 2.5 cloud agents from review.

Queue Cursor Agent SDK fix handoffs on Composer 2.5 after Critique review — harness deep dive, benchmarks, and Settings BYOA setup.

Open

Skills

Agent Skill Marketplace.

Publish and version critique-review lenses — Fintech Security, Kafka migrations, agentic audits — and import them into any Critique workspace or agent runtime.

Open

Agent API

Best cloud coding agent API (2026).

HTTP runs on OpenCode + E2B — crt_ keys, draft PRs, warm sessions, cookbooks, and pricing from $19/mo shared credits or OpenRouter BYOK.

Open

Compare

Critique vs Devin API.

Cloud coding agent API hub — model freedom, OpenRouter :free routes, pay-as-you-go credits, and when Devin’s session UI fits better.

Open

Deep dive

Persistent Coding Agent API sessions.

Keep OpenCode and the E2B sandbox warm between turns on one run id — multi-step automation with SSE and explicit teardown.

Open

Ship log

What shipped lately.

June 2026 (v6.5.0): v6 adds Platform API write paths, lifecycle webhooks, and a lighter default theme — full outcome bullets on /version.

Open

Essays

Read the Critique blog.

Field notes on multi-model review, hybrid retrieval, operational trust gaps, and how we ship the product.

Open

Engine

Every model we route.

20+ frontier and mid-tier models — GPT-5.4, Claude Opus 4.7, Gemini 3 Pro, Kimi K2.6, GLM-5 — swappable as lead or sub-agent.

Open

Docs

Install guide and API reference.

GitHub App install, policy tuning, Remedy fix-agent, chat lane, webhooks, and SSO for enterprise.

Open