Deploy and Host Omnigent on Railway

Omnigent is an open-source control plane for AI coding agents. It keeps session history, artifacts, accounts, the agent catalog and an MCP proxy with server-side policy enforcement in one place, so a team shares context instead of each person running an agent alone on a laptop. It is agent-agnostic: Claude Code, Codex and others run under it without being rewritten.

This template deploys the Omnigent server together with a managed PostgreSQL, already wired to each other, with your admin account created on first boot.

About Hosting Omnigent

The server is a FastAPI application backed by PostgreSQL. Three things have to be right for a deployment to work and to stay safe, and this template handles all three.

The database connection is injected as a Railway reference rather than a copied connection string, so no password is ever written into the configuration. Artifacts, bundles and uploads are kept on a persistent volume mounted at /data/artifacts; without one they sit on the container's ephemeral disk and disappear on every redeploy.

The third is less obvious. A brand-new Omnigent exposes an unauthenticated first-run setup endpoint — it has to, or setup could not be completed. A public instance reached before you open that form can be claimed by whoever arrives first, and the generated hostname is not a secret: it appears in public certificate transparency logs within minutes. This template closes the window by creating the admin during the first boot, so by the time the URL answers, setup is already complete.

Why Deploy Omnigent on Railway?

Railway runs the server and the database side by side on a private network, with HTTPS and a generated domain out of the box, so there is nothing to configure for the two to find each other.

The image is pulled straight from the Omnigent project's own registry and pinned to a specific release, so there is no build step to wait for or to fail, and two people deploying a week apart get identical software. Omnigent publishes daily development builds, which is exactly why an unpinned tag would be a poor choice here.

Common Use Cases

A shared, self-hosted hub where a team runs and watches AI coding agents, with history and artifacts kept in a database you control.
A policy and collaboration layer in front of several agent harnesses, so switching between them does not mean rebuilding your workflow.
A private Omnigent instance for work that cannot be sent to a third-party service, with accounts and session logs staying on your own infrastructure.

Dependencies for Omnigent Hosting

Nothing external is required to stand the server up: PostgreSQL is included in this template.

To actually run agents you register a host yourself. Sessions execute wherever omni host is running — by default your own machine, with its files and credentials — so a session only runs while that machine is awake. Cloud runners such as Modal, Daytona and Blaxel are configured separately and are not part of this template.

omni login  https://YOUR-DEPLOYMENT.up.railway.app
omni host   https://YOUR-DEPLOYMENT.up.railway.app

Deployment Dependencies

Omnigent — https://github.com/omnigent-ai/omnigent (Apache-2.0)
Official server image — ghcr.io/omnigent-ai/omnigent-server
PostgreSQL — https://www.postgresql.org

After deploying

You are asked for an admin username during deployment. The password is generated for you and can be read from the omnigent service's variables, under OMNIGENT_ACCOUNTS_INIT_ADMIN_PASSWORD. Self-service registration is invite-only out of the box, so invite teammates from the Members screen.

To move to a newer Omnigent release, open the omnigent service, then Settings → Source, change the image tag and redeploy.

This is a community template and is not affiliated with the Omnigent project.