Deploy and Host OpenHands Agent Canvas on Railway

OpenHands is an open-source AI software-development agent — the self-hosted alternative to Devin. It writes code, runs commands, edits files and browses the web on your behalf, from a browser UI. This template deploys Agent Canvas 1.14.0, the current OpenHands application server, as a single Railway service with a persistent volume.

About Hosting OpenHands Agent Canvas

Agent Canvas ships as one image containing the agent server, the automation server and the web UI behind a single port. This template runs that image with three Railway-specific changes:

The session API key is never written into the served page. Stock Agent Canvas embeds its key in the HTML at /canvas/ so that a browser on the same machine can pick it up automatically. On a public URL that hands the key — and therefore arbitrary shell access inside the container — to anyone who opens the page. This build starts the server in --auth-required mode instead, so the UI asks you for the key and the API rejects every unauthenticated request.
The volume is made writable before the app starts. Railway mounts volumes owned by uid 0, while the image runs as uid 10001. Without a repair the automation database fails to open and the container exits during startup.
Telemetry is off. DO_NOT_TRACK=1 is baked into the image.

The image is prebuilt and pinned to 1.14.0, so a deploy is an image pull rather than a source build, and a later redeploy does not silently upgrade the agent.

Why Deploy OpenHands Agent Canvas on Railway?

Railway gives the agent a persistent disk for its conversations, workspaces and stored LLM credentials, a public HTTPS URL, and a one-click redeploy. The session key and the secret that encrypts stored credentials are both generated per deployment, so nothing sensitive is shared between installs of this template.

Common Use Cases

A hosted coding agent you can reach from any machine, without running Docker locally.
Long-running refactors, migrations and test-suite fixes that outlive a laptop session.
Driving the agent over its HTTP API from your own scripts or automations.

Dependencies for OpenHands Agent Canvas

An API key for the LLM provider you want the agent to use (OpenAI, Anthropic, or any LiteLLM-supported provider). You enter it in the UI after the first sign-in; it is stored encrypted with this deployment's own OH_SECRET_KEY.

Deployment Dependencies

Upstream project: https://github.com/OpenHands/OpenHands
Upstream image: ghcr.io/openhands/agent-canvas:1.14.0
Railway wrapper image: https://github.com/bon5co/openhands-railway

First steps

Deploy, then open the service URL. You land on the key-entry screen.
Copy OH_SESSION_API_KEYS_0 from the service's Variables tab and paste it in.
Add your LLM provider key in Settings, then start a conversation.