Skip to content
AICodeReviewer

Quick start

This guide walks through the two fastest ways to run AICodeReviewer (AICR): Docker Compose for a deployment-like setup, and local Node.js for development. Pick one, then verify with a health check and a dry-run review.

Docker Compose

Section titled “Docker Compose”

The example/ directory ships a ready-to-edit Compose stack.

Terminal window
cd example/


# 1. Create .env from the sample and fill in secrets
cp .env.sample .env
# Edit .env: set AICR_LLM_API_KEY, AICR_GITEA_TOKEN, AICR_WEBHOOK_SECRET


# 2. Edit config.yaml: set your Gitea URL, repo, and model


# 3. Build and start
docker compose up -d

The Compose file mounts config.yaml read-only and persists three named volumes: aicr-data (queue DB, run history), aicr-workspaces (cloned repo caches), and aicr-logs. It exposes the service on port 8080 and ships a health check against /healthz.

What goes where:

  • example/.env — every secret as an environment variable. Never commit this file. See Authentication & secrets.
  • example/config.yaml — non-secret configuration. References env var names (e.g. api_key_env: AICR_LLM_API_KEY), never values.
  • example/docker-compose.yaml — the stack definition.

Local Node.js

Section titled “Local Node.js”

For development or evaluation without Docker:

Terminal window
# From the repository root:


# 1. Install dependencies and build runtime packages
pnpm install
pnpm build


# 2. Set environment variables (one-time source, or export manually)
source example/.env


# 3. Start the server
node packages/cli/dist/index.js serve \
  --config example/config.yaml \
  --port 8080

Verify the server is running

Section titled “Verify the server is running”

Once the server is up, confirm it answers the health endpoint:

Terminal window
curl http://localhost:8080/healthz
# ok

The /healthz endpoint returns a plain-text ok and is the canonical liveness probe (also used by the Compose health check). If you enabled the observability dashboard (admin.* in config), visit http://localhost:8080/dashboard.

Dry-run a review

Section titled “Dry-run a review”

Before wiring up a webhook, run a one-shot review against a local checkout to confirm the LLM and pipeline are healthy. Dry-run does not publish any output.

Terminal window
export AICR_LLM_API_KEY=sk-xxx


node packages/cli/dist/index.js review \
  --config example/config.yaml \
  --repo "my-org/my-repo" \
  --provider gitea \
  --source-root . \
  --dry-run

Flags of note:

  • --provider — the trigger provider kind from your config (gitea, github, gitlab, p4, svn).
  • --source-root — the checkout to review.
  • --dry-run — prepare and run the review but skip all output channels.

If the dry-run succeeds, the LLM credentials, agent, and sandbox are wired up correctly. Next, point a VCS webhook at the server — see Your first webhook review.

Next steps

Section titled “Next steps”