Deploy and Host Hermes Agent on Railway

Hermes Agent is Nous Research's self-improving AI agent: it builds skills from experience, refines them as it works, searches its own past conversations, and keeps a persistent model of you across sessions. It is the open-source, self-hosted answer to OpenClaw and the other hosted agent products — the same always-on assistant, running on infrastructure you own, with no per-seat pricing and nobody else holding your agent's memory.

This template deploys it with nothing to configure. The deploy form has a single field, the dashboard password, and it arrives pre-filled with a generated secret. There is no port to pick, no admin username to invent, no data path to type. Click deploy, then sign in.

About Hosting Hermes Agent

Three things decide whether a self-hosted Hermes is any good, and all three are settled here before you click deploy.

Memory has to survive. The whole premise of Hermes is continuity — it is supposed to remember what it learned and get better at your work over time. All of that lives under HERMES_HOME at /opt/data: learned skills, session history, identity, configuration. Deploy it without a volume and you get an agent with amnesia that rebuilds nothing and remembers no one. This template mounts a volume there, so the agent that exists next week is the one you trained this week.

The dashboard has to be gated, and it fails closed. Hermes ships an auth gate that engages automatically on any non-loopback bind, and it refuses to start at all without a registered auth provider. As of the June 2026 hardening, HERMES_DASHBOARD_INSECURE no longer disables that gate — it is accepted and ignored — because unauthenticated public dashboards were the entry point for a real credential-persistence campaign. So an empty password is not "an open dashboard", it is "no dashboard". This template registers the bundled password provider with an admin user and a generated password; if you clear the field entirely, the image generates one, persists it on the volume, and prints it to the deploy logs rather than crash-looping.

The listening port has to match Railway's. Railway assigns the port your service is reached on, and the dashboard follows it — so the public URL works on the first deploy instead of returning Application failed to respond.

Rather than ask you to supply those settings, they are baked into the image this template deploys: a thin, open wrapper around Nous Research's own published build (nousresearch/hermes-agent), rebuilt weekly to track upstream releases. Hermes runs under s6 supervision inside the container, which is what keeps the dashboard alive independently of the agent process.

Why Deploy Hermes Agent on Railway?

Railway is a singular platform to deploy your infrastructure stack. Railway will host your infrastructure so you don't have to deal with configuration, while allowing you to vertically and horizontally scale it.

By deploying Hermes Agent on Railway, you are one step closer to supporting a complete full-stack application with minimal burden. Host your servers, databases, AI agents, and more on Railway.

Nothing to configure — one pre-filled field on the deploy form, and it is the password. No port, no username, no data path.
Memory that persists — skills, sessions, and identity on a volume, which is the entire point of a self-improving agent.
A dashboard that cannot come up unauthenticated — a real password provider is registered, not an auth bypass.
Official upstream image — Nous Research's own published build, wrapped for Railway, not a re-implementation.
Always reachable — the agent runs on infrastructure, so you can talk to it when your laptop is closed.
Any model you like — Nous Portal, OpenRouter, OpenAI, or your own endpoint; no lock-in.

Common Use Cases

An OpenClaw alternative you own — the same always-on agent workflow, self-hosted, with your data and its learned skills staying on your volume.
A personal agent that compounds — it learns your workflows and keeps the skills it builds, session after session.
Long-running background work — hand off research or a multi-step task and check the dashboard later.

Dependencies for Hermes Agent Hosting

A volume mounted at /opt/data for agent memory and configuration (included in this template).
An API key for a model provider (Nous Portal, OpenRouter, OpenAI, or a compatible endpoint). This is the one thing the template cannot supply for you — add it from the dashboard after deploy, before giving the agent real work.

Deployment Dependencies

Hermes Agent — upstream project by Nous Research
nousresearch/hermes-agent — the official image this build is based on
bon5co/hermes-agent-railway — the open Railway wrapper this template deploys, so you can read exactly what was changed

Implementation Details

One service, one volume, one variable:

Service	Image	Volume	Public
`hermes`	`ghcr.io/bon5co/hermes-agent-railway:latest`	`/opt/data`	yes

Variable	Default	Why it is the only one
`HERMES_DASHBOARD_BASIC_AUTH_PASSWORD`	generated secret	It is the one value that must be yours. Everything else — port, dashboard bind, admin username, `HERMES_HOME` — is baked into the image.

After deploy, open your public domain and sign in as admin with that password, then change it to your own once you are in. Then add a model provider key from the dashboard — Hermes needs one to do anything beyond starting up.

Do not delete the volume. It holds every skill the agent has learned and every session it remembers; there is nothing to restore it from.