Platforms

Custom Platform

Run Rivet workers on any platform that can run your app.

A worker is your app plus two environment variables. Any platform that can run your app and reach the control plane can host it.

Requirements

Your platform must provide:

A way to run your app, as a container, a process, or a serverless function.
Environment variables, to carry the connection.
Outbound network access to the control plane.
For serverless platforms only, a public HTTP endpoint so the control plane can reach /api/rivet.

Steps

Pick a runtime mode

Runner, the default, for anything that keeps a process alive: containers, VMs, managed app platforms. The worker dials out, so it needs no public URL.
Serverless, set with RIVETKIT_RUNTIME_MODE=serverless, for request-driven platforms with no long-lived process. The control plane calls in, so the URL must be public.

See Workers for how the two differ.

Deploy your app

Deploy however your platform expects. If you are in serverless mode, confirm /api/rivet is publicly reachable and returns a valid response.

Set the environment variables

RIVET_ENDPOINT=https://<namespace>:<secret-token>@<control-plane-host>
RIVET_PUBLIC_ENDPOINT=https://<namespace>:<publishable-token>@<control-plane-host>

RIVET_PUBLIC_ENDPOINT is served to browsers. Never put the secret token in it.

Connect

In runner mode there is nothing to register: the worker appears under Runners in the dashboard once it starts.

In serverless mode, select Custom as your provider and paste your URL with the /api/rivet path into the connect form, for example https://my-app.example.com/api/rivet.

Configure graceful shutdown

Actors drain for up to 30 minutes. Give the process at least 35 minutes between SIGTERM and SIGKILL, and make sure the signal reaches it: a wrapper such as npm start becomes PID 1 and swallows it.

Troubleshooting

The worker never appears under Runners. In runner mode, check the control plane logs first. A no_runner_config rejection there means no runner config exists for the pool, which the worker reports only as a silent reconnect loop. See Runner configs. Otherwise check outbound network access and that RIVET_ENDPOINT carries the right namespace and token. In serverless mode, check that the registered URL is reachable from the public internet.

Actors do not respond. Confirm both RIVET_ENDPOINT and RIVET_PUBLIC_ENDPOINT are set. Missing the public one leaves clients with no connection details.

The connection drops every 30 to 60 seconds. A proxy in front of your worker is timing out idle WebSockets. Raise its idle timeout to 3600 seconds. See Ports.