Quickstart

Run auto-swe on your own machine and start a first workflow, from a clean checkout to a run you can watch in the dashboard.

This is the local-development path. For a production install, follow deployment.md instead.

What you need

• Node.js 26 or later and Docker with Docker Compose.
• An Anthropic API key and an OpenAI API key. The seeded agents use Anthropic models, and semantic memory embeds with OpenAI. Both are added in the dashboard, not in .env — see model-configuration.md to use other providers instead.
• A GitHub token with access to a repository you are happy to receive a pull request on, for the engineering workflow. Workflows that do not touch a repository do not need one.

1. Install

npm install -g corepack && corepack enable
yarn install

Node 26 no longer bundles Corepack, which is why it is installed from npm first.

2. Configure the environment

yarn env:setup init

This creates .env from .env.example and generates every required secret. It never overwrites an existing .env; after pulling a change to .env.example, run yarn env:setup sync to bring your file up to date (it keeps your values and backs the old file up first), and yarn env:setup check to validate it. The values it generates:

Variable	What it is
CONFIG_ENCRYPTION_KEY	Base64 of 32 random bytes. Encrypts every stored credential; the gateway and worker refuse to start without it. Never regenerated once set — rotate it with yarn keys:rotate
SEED_ADMIN_PASSWORD	The password for the seeded admin account, admin@auto-swe.local
BETTER_AUTH_SECRET, JWT_SECRET	Session-cookie and API-token signing keys
GARAGE_RPC_SECRET	Node auth for the bundled Garage object store (only while COMPOSE_PROFILES includes objectstore)

The generated admin password is in .env; the script never prints secrets.

Model keys, GitHub, Slack, and every other integration are stored encrypted in the database and managed from the dashboard. GITHUB_TOKEN in .env still works as a bootstrap fallback.

3. Start the infrastructure

yarn docker:infra:up

This starts PostgreSQL (with pgvector), a second PostgreSQL for Temporal, and the Temporal server and its web UI — plus the Garage object store, because .env.example enables its objectstore profile.

4. Create the database

yarn db:migrate && yarn db:generate && yarn db:seed

The seed creates the admin account, a default team, the built-in workflow templates, skills, scanner patterns, and agents.

5. Start the gateway and the dashboard

In two terminals:

yarn dev:gateway   # http://localhost:8080
yarn dev:web       # http://localhost:3000

Not the worker yet: it checks at boot that every agent can reach a model, and refuses to start until the credentials below exist.

6. Add model credentials

Sign in at http://localhost:3000 as admin@auto-swe.local, then open Build → Model configuration (/models) and add a credential for anthropic and one for openai on the Credentials tab.

7. Connect GitHub and a repository

• Admin → Integrations → GitHub (/admin/integrations): enter the token and save. Skip this if GITHUB_TOKEN is set in .env.
• Connections (/connections): add the repository as a Git repo connection, or import it from GitHub.

8. Start the worker

yarn dev:worker

If anything is still missing, it exits with one error listing every missing piece and where in the dashboard to add it.

9. Start a run

In the dashboard, choose + New request, pick the ticket-to-merge template, choose the repository, and describe a small change — “add a GET /health endpoint that returns { status: 'ok' }” is a good first one.

Then watch it:

Where	Shows
Runs (/runs)	Every run, and per run its steps, agent transcripts, tool calls, and cost
Work → Approvals (/approvals)	Steps waiting for a person, in workflows that have them
Temporal UI, http://localhost:8233	The durable workflow underneath: its event history and signals

The run ends with a pull request on your repository and then waits for a person to merge it. By default auto-swe never merges (the opt-in auto mode of pr-evaluation.md is off).

Next

• concepts.md — the vocabulary: runs, templates, nodes, gates, agents, skills, scopes.
• architecture.md — how a run moves through the system.
• The CLI (packages/cli/README.md) and the REST API start the same runs headlessly, with a personal access token from Settings → API tokens.