AgentOSDeploy AgentOS on Railway with the runtime and data boundaries understood.
The published AgentOS template is the supported Railway path. It combines the application and a pinned OpenClaw Gateway in one service, uses a persistent volume, and keeps the Gateway on container loopback.
1. Before deployment
Use the published Railway template for the supported hosted path. You will need a Railway account, an initial administrator password, and a model-provider key before real work can run. Railway infrastructure and selected model-provider usage are billed separately.
2. Deploy the published template
The AgentOS Railway path runs AgentOS and a pinned OpenClaw Gateway in one service. The template provisions a persistent `/data` volume and generates internal AgentOS and OpenClaw tokens. The only required operator input is the initial administrator password.
3. Protect and sign in
Open the generated Railway domain and sign in with the initial administrator credentials. The initial password creates the protected account only when no Instance Protection state exists; changing the deployment variable later does not reset the account.
4. Verify the deployment
Use Railway’s deployment and application logs to confirm the service starts and its healthcheck passes. Railway healthchecks validate deployment readiness; they are not continuous monitoring. A service with an attached volume can have brief downtime during redeployment.
5. Preserve the runtime state
The published template uses `/data` for persistent AgentOS and OpenClaw state across redeploys. Do not remove or replace the volume without understanding the data you are discarding. Railway volumes are mounted at runtime, not during build or pre-deploy steps.
6. Connect a ready model
Use AgentOS Setup Center to connect a provider and verify that a model is ready. AgentOS uses your provider keys; model usage is billed by the provider, separately from Railway infrastructure.
7. Create the first operating loop
Create a workspace, configure one digital worker with a clear role and context, assign a scoped task, then inspect the output and runtime state before adding schedules or more access.
8. Update deliberately
Use the project’s update preflight and compatibility information before updating OpenClaw. AgentOS currently expects OpenClaw 2026.6.8 or newer; run `agentos doctor --deep` when preparing a trusted local environment. For Railway updates, review deploy logs and retain the volume.