Deploy and Host DeepSeek Harness on Railway

DeepSeek Harness is DeepSeek's open-source coding agent. It reads and edits code, runs commands and works through tasks with you from a browser. This template runs its web UI as a Railway service behind a password, with sessions and repos kept on a volume.

This is a community template, not made or supported by DeepSeek.

About Hosting DeepSeek Harness

The web UI refuses to listen on anything but loopback and logs in with a one-time token printed to the console, which doesn't suit a server. This template puts Caddy in front on Railway's public port. You sign in with WEB_USER and WEB_PASSWORD, and the proxy passes DSH's token on for you, so you never handle it and it never reaches the deploy logs.

Settings is usually disabled on any address other than localhost. The image patches that one check so the Models page works from your Railway domain, and the build fails if a new DSH version changes that code.

After deploying, open the service's domain and sign in. Set DEEPSEEK_API_KEY, or save a key under Settings > Models from the account menu at the bottom left. Then choose Add workspace, pick the workspace folder, and start a session. Without a key the UI opens but the agent can't answer. Model usage is billed by your provider, not by this template.

The agent has full shell access to this container. Anyone who has the password can run commands in it and read its variables, including GH_TOKEN and your API key. Use a long password and a token limited to the repos the agent should touch.

Common Use Cases

Run a coding agent that keeps working after you close your laptop
Use the same agent and sessions from a phone, tablet or a second computer
Give the agent its own machine to clone, build and test in, away from your own files

Dependencies for DeepSeek Harness Hosting

A DeepSeek API key or another model provider's key
Optionally, a GitHub token for private repos and pushing

Deployment Dependencies

Implementation Details

Volume: /data is the home directory. Sessions and settings live in /data/.dsh, and /data/workspace is there for your projects. All of it survives redeploys, including files the agent wrote.
Tools included: git, git-lfs, gh, ripgrep, tmux, Python 3, Node.js 24 and build-essential. Packages the agent installs with apt go away on redeploy; anything under /data stays.
Memory: plan on at least 1 GB. The UI idles at about 300 MB and runs on Railway's 0.5 GB Free limit, but in our test cloning a small Node project and running npm install there killed the whole service. It restarts by itself with your files and sessions intact, though the running turn is lost. The same job, including its test suite, finished on a 1 GB limit. Larger projects will need more.
Other providers: Settings > Models takes a custom Anthropic-compatible base URL and model names, or a provider from the built-in catalog under Add model provider.
Version: pinned to a tested release, and upstream is still publishing release candidates, so expect its interface to change between updates.

Why Deploy DeepSeek Harness 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 DeepSeek Harness 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.