Self-Host

axon build produces a Docker image from your agent folder. Deploy it anywhere that runs containers. You manage the runtime.

axon build
# → .agent/image.tar

The image contains your agent source, installed modules, and resolved dependencies. Your .env file is excluded — pass secrets as environment variables at runtime.

What self-hosting means

Your agent — the source, the tools, the data, the container — runs on your infrastructure. The cognitive runtime (the loop, context assembly, tool dispatch) is managed by Axon and runs on ArcLabs servers. Your container connects to it at boot.

This means your agent's data never leaves your infrastructure. The cognitive work happens on Axon's side, but nothing is stored there — it's stateless by design.

The agent expects one environment variable

AXON_API_KEY=axon_...   # your Axon API key — required to connect to the cognitive runtime

Everything else your agent needs (e.g. GITHUB_TOKEN) goes in as additional env vars, matching what you declared in axon.config.ts.

Provider guides

Pick the platform that suits you:

• Railway — simplest. Push image, set env, done.
• Fly.io — good for persistent workloads and global edge.
• Render — straightforward, generous free tier.
• Google Cloud Run — scales to zero, pay per request. What Axon Cloud runs on.

All four follow the same pattern: load the image, set environment variables, expose port 3000.