Getting Started
Connect Critique to GitHub, run your first review, and learn where to configure policy and credits.
This guide gets you from zero to a working PR review in a few steps.
1. Create an account
Sign up at critique.sh/sign-up with email and password, or use GitHub or Vercel OAuth. Sessions last about a week and refresh automatically when you are active.
Signing in with GitHub is recommended if you will install the Critique GitHub App — it keeps account linking straightforward.
2. Install the GitHub App
Critique reviews code through a GitHub App installation, which is separate from “Sign in with GitHub”:
- Open the dashboard.
- Follow Connect GitHub (or go through
/github/setupafter installing from GitHub). - Choose which organizations and repositories Critique can access.
- After install, you land on the dashboard with
connected=github.
You can add or remove repositories anytime from GitHub’s app settings or from Automation in the dashboard.
3. Turn on automation
Open Dashboard → Automation for the installation you care about:
- Confirm repositories are synced and show as active.
- Set review policy (strictness, which specialists run, optional model overrides). Defaults are strict enough for most teams; see Policy fields.
- Optionally configure repository secrets when reviews need environment variables (same idea as CI secrets — never paste secrets into PR comments).
When a pull request is opened, updated, or reopened, Critique queues a review automatically.
4. Open a pull request
Push a branch and open a PR on a connected repository. Within a short time you should see:
- A check run on the PR (Pass / Warn / Fail depending on findings and policy).
- A summary comment with the verdict and a link to the full run in Critique.
- Inline annotations on lines where findings map to the diff.
You can also type @critique /review on any PR to re-run manually. See @critique bot.
5. Inspect results in the dashboard
Dashboard → Review runs lists recent pipelines. Open a run to see:
- Findings grouped by specialist and severity.
- Execution timeline and model usage.
- Remedy, Queue agent (Cursor / Claude / Codex BYOA), Fix prompt, or handoff JSON when you want to act on feedback.
Shareable links from GitHub comments point to the same run (sign-in required for full detail).
6. Understand credits
Reviews, Remedy, chat, and indexing consume credits from your plan’s monthly pool. The dashboard Usage page shows what ran and what remains.
- Compare plans on Pricing.
- Read Billing & credits for how pools, floors, and BYOK interact.
OpenRouter BYOK and BYOA agents
Settings supports BYOK and BYOA:
- OpenRouter key (BYOK — you pay OpenRouter for review/chat models)
- Cursor, Anthropic, and OpenAI keys (BYOA — queue Cloud Agents, Claude Managed Agents, or Codex from completed review runs; execution bills your vendor accounts)
Add keys after your first review, then use Queue agent on any completed run — or stay with one-click Remedy inside Critique.
What to read next
| If you want to… | Read |
|---|---|
| Tune merge blocking vs warnings | Policy fields |
| Filter spammy or low-trust PRs | Checkpoint |
| Ask questions about the repo | Chat & workspace — or open /workspace directly. Launch essay: Introducing Critique Workspace. |
| OpenRouter BYOK for model spend | BYOK |
| Cursor, Codex, Claude Code, Copilot for fixes | BYOA |
| See how async processing works | Event-driven pipeline |