onauto/runtime
Auto Runtime. Build, run and manage reusable playbooks and workflows with your AI agent.
10K+
onauto/runtime repository overview
Auto Runtime
Auto Runtime runs the playbooks your agent builds: their reasoning, interaction, prediction, recall and computation functions, and the workflows that coordinate them. Your coding agent connects over MCP to build and run them, and Auto Studio lets you inspect and manage what it built.
Documentation · Self-hosting guides · Source on GitHub · Release notes
Quick start
A person signs in through the Auto identity service, and the runtime needs only the address it is reached at, PUBLIC_URL. Put it in a file of settings, and start the server with its port published on this machine's loopback address and a volume for its ledger:
echo 'PUBLIC_URL=http://localhost:5555' > auto-runtime.env
docker volume create auto-runtime-data
docker run --detach --name auto-runtime --restart unless-stopped \
--publish 127.0.0.1:5555:5555 --env-file auto-runtime.env --volume auto-runtime-data:/data \
docker.on.auto/onauto/runtime:latest
curl http://localhost:5555/health
Then connect your agent to /mcp, for example in Claude Code, and sign in when it opens the Auto identity service in your browser:
claude mcp add --transport http auto-runtime http://localhost:5555/mcp
Run with Docker walks through each step, adds a model provider, gives a machine or a pipeline an API key and has a Compose file that runs the runtime beside PostgreSQL.
What a deployment needs
| Need | What to use |
|---|---|
| The ledger | A volume on /data, where it keeps a SQLite file, or a PostgreSQL database named by DATABASE_URL: Use PostgreSQL |
| Secrets key | SECRETS_KEY, 32 random bytes in base64 that seal the secrets kept in the ledger: Deploy for your team |
| Sign-in | PUBLIC_URL, the public https address people reach the runtime at: Deploy for your team |
| API keys | API_KEYS, for machines and pipelines, with an entry for each key the key command makes: Authentication and security |
| A model | Anthropic, OpenAI, Google, Amazon Bedrock, Azure OpenAI, Vertex AI or your own gateway: Model providers and gateways |
| HTTPS | A reverse proxy or load balancer in front of port 5555: Deploy for your team |
| Object storage | Optional, an S3-compatible bucket for large content: Keep content on object storage |
Every setting is an environment variable; The configuration reference lists them all, with the configuration file for tool servers, gateways and model aliases. Troubleshooting covers the common problems.
Pull the image through docker.on.auto/onauto/runtime: it is this same image, and Scarf counts each pull for Auto. When it starts, the runtime sends one anonymous usage ping: its version, platform, processor, Node.js version, that it runs from the image, and a random install id kept on the /data volume, never anything of its work. DO_NOT_TRACK=1 turns it off. Usage ping lists exactly what it sends.
Tags
| Tag | Points to |
|---|---|
X.Y.Z, such as 0.9.1 | Exactly that release; it never changes |
X.Y, such as 0.9 | The newest release of that line |
latest | The newest release |
Every tag is published for linux/amd64 and linux/arm64, built in GitHub Actions with a software bill of materials and provenance. The image runs as the unprivileged user node, checks its health at /health and stops cleanly on SIGTERM.
Without Docker
npx @on.auto/runtime runs the same server on macOS or Linux with Node.js 26.10 or later, on http://localhost:5555, in local mode, where every request on that computer acts as you, with no sign-in. See the local quick start.
License
Source-available under the Elastic License 2.0.
Tag summary
Content type
Image
Digest
sha256:55716d653…
Size
127.5 MB
Last updated
12 days ago
docker pull onauto/runtime