Deploy and Host OpenCode on Railway

OpenCode — open-source AI coding agent

OpenCode is the open-source AI coding agent (think Claude Code, but open and provider-agnostic) — it reads your repo, writes and edits code, runs commands, and opens pull requests. This template runs it as a Railway service with two front doors: a browser web UI on a public domain, and the opencode terminal UI over railway ssh — both sharing one /workspace, auth, and repos that persist on a volume.

About Hosting OpenCode

One container runs opencode web on a public, password-protected Railway domain; railway ssh into the same container gives you the opencode TUI — the real terminal agent, not a web console imitating one. A /workspace volume keeps your repos and OpenCode credentials across redeploys, and git + the GitHub CLI are preinstalled for clone / commit / PR work. Railway provisions TLS, the domain and the volume, and restarts on failure.

Deploys come from a prebuilt, version-pinned image, so there is no build to sit through and every deploy runs the exact OpenCode build that was tested — new deploys pick up the latest patched image automatically, while a box you already have keeps running undisturbed until you redeploy it. The web UI password is generated for your deploy; you never invent one or inherit a shared default. Add a provider key on the deploy form, or run opencode auth login after connecting.

Why Deploy OpenCode on Railway?

A real terminal, not just a web console — railway ssh gives you the actual opencode TUI with a shell, git, and gh beside it. Most hosted OpenCode setups stop at the browser.
Two front doors, one service — a shareable browser web UI and that terminal, both driving the same workspace, auth, and sessions.
Deploys in seconds, pinned to a tested build — a prebuilt image means no build to wait through and no surprise from an upstream release published an hour ago. Your running box is never restarted out from under a session; you upgrade when you choose to.
Your laptop can sleep — the agent lives in the cloud; reopen the web link or railway ssh back in and pick up where you left off.
Secure by default — the public web UI sits behind HTTP basic auth with a password generated uniquely for your deploy; SSH is Railway-authenticated.
Persistent — repos and OpenCode auth sit on a volume, so a redeploy never logs you out or loses your working tree.
Provider-agnostic — bring an Anthropic, OpenAI, or OpenRouter key (or opencode auth login to anything OpenCode supports); switch models without rebuilding.

Common Use Cases

A cloud coding-agent workstation reachable from a browser link or the terminal.
Long-running refactors or migrations that keep going regardless of your local machine.
A shared, reproducible agent environment with pinned tooling across repositories.

Dependencies for OpenCode Hosting

A provider API key for at least one model backend — ANTHROPIC_API_KEY, OPENAI_API_KEY, or OPENROUTER_API_KEY, all three offered on the deploy form so the box is usable the moment it boots. Leave them blank and run opencode auth login after connecting instead.
The Railway CLI for terminal/TUI access (railway ssh); the web UI needs only a browser.

The box also honours GITHUB_TOKEN, GIT_USER_NAME and GIT_USER_EMAIL if you add them under Variables later — GITHUB_TOKEN authenticates gh and HTTPS git on boot. They are kept off the deploy form on purpose: the box generates its own SSH key on first boot (cat ~/.ssh/id_ed25519.pub → add it to GitHub), which is a safer default than pasting a personal access token into a form.

Deployment Dependencies

OpenCode — the open-source AI coding agent (opencode-ai on npm).
GitHub CLI — bundled so the agent can manage issues and pull requests.

Why This Template?

One click gives you OpenCode with two front doors — a browser web UI and an SSH terminal — over a single service, with credentials and repos that outlive every redeploy. Source and docs: github.com/yuting1214/opencode-railway.