DSH Coding Agent with Web UI

Deploys DSH (dsh), the open-source agent harness built on DeepSeek Harness, release 0.2.1-rc.2, as an always-on coding agent with a browser UI. A Caddy proxy puts a generated password in front of it, and sessions, settings and workspaces live on a volume. You supply a model provider key.

What you get

  • One service running the dsh Web UI on loopback behind a Caddy password gate, with healthcheck /healthz
  • A generated DSH_WEB_PASSWORD (user DSH_WEB_USERNAME, default admin); the Harness's own session cookie then lasts 30 days
  • A Railway volume at /data for settings, credentials, sessions, plugins and workspaces
  • New sessions use upstream's workspace-write preset: writes limited to the session workspace, approval prompts for anything wider
  • OpenTelemetry upload off by default

Good for

  • A coding agent reachable from any browser or phone that keeps working while your laptop is closed
  • Repository summaries, refactors, test runs and dependency upgrades on code cloned into the workspace
  • Scheduled reminders and recurring checks with the Harness scheduler
  • Trying plugins, agent presets and different model providers without a local install

Services in this template

  • deepseek-harness · from baranberkay96/dsh-railway

Questions people ask

How do I log in?

Open the public URL, sign in with DSH_WEB_USERNAME and DSH_WEB_PASSWORD from the service Variables. If you did not set DEEPSEEK_API_KEY, add a key under Settings, Models, or add another provider. Then pick a workspace and start a session.

What are the security and privacy considerations?

The agent can run shell commands and spend API credits, so keep the password private and set spending limits at your provider; read upstream's SAFETY.md first. Upload of session logs to DeepSeek is on by default when you use the official DeepSeek API; switch it off in Settings, General if you prefer.

Is data persistent?

Yes. Settings, credentials, sessions and workspaces are on the /data volume. The Harness is a single-writer process with file-based state, so keep one replica and scale vertically.

How do I use a custom domain or upgrade?

List the domain in DSH_TRUSTED_HOSTS and redeploy; the Harness refuses API calls from untrusted host names. To upgrade, change DSH_VERSION in the Dockerfile and redeploy. The project is a developer preview with breaking changes between releases, so read the release notes first.

What is the licence?

DeepSeek Harness is MIT-licensed. This is a community template, not affiliated with or endorsed by DeepSeek. Model usage is billed by your provider.

Full template README

Deploy and Host DSH with Railway

Run DSH (dsh), the open-source agent harness built on DeepSeek Harness, as an always-on coding agent with its browser UI. It reads and edits files, runs commands, delegates to subagents, schedules tasks and loads plugins, while sessions, settings and workspaces stay on a Railway volume. This is a community template, not affiliated with or endorsed by DeepSeek.

About Hosting DSH

Upstream runs the Web UI with npx @deepseek-ai/dsh web on 127.0.0.1:3080, so it lives only as long as your laptop does. This template installs the pinned npm release 0.2.1-rc.2 in one container and keeps the Harness on loopback, following upstream's reverse-proxy guide. A Caddy reverse proxy in the same container owns the public port: it asks for a generated password first, then hands your browser the Harness's own login token, so you sign in once and the Harness's signed session cookie keeps you signed in for 30 days, across restarts. Railway's edge supplies HTTPS. The agent runs as an unprivileged user with upstream's sandbox and approval prompts.

Common Use Cases

  • A coding agent you can reach from any browser or phone that keeps working on a repository while your laptop is closed
  • Repository summaries, refactors, test runs and dependency upgrades on code cloned into the persistent workspace
  • Scheduled reminders and recurring checks with the Harness's built-in scheduler
  • Trying DSH plugins, agent presets and different model providers (DeepSeek, Anthropic, OpenAI, OpenRouter, Kimi, GLM or any compatible gateway) without a local install

Dependencies for DSH Hosting

  • A model provider key: a DeepSeek API key for the default model (set DEEPSEEK_API_KEY or enter it in Settings → Models), or any other supported provider added in the Web UI
  • A Railway volume at /data (included) for settings, credentials, sessions, plugins and workspaces

Deployment Dependencies

Implementation Details

Service Source Purpose
deepseek-harness Dockerfile in services/deepseek-harness (Node.js 24, @deepseek-ai/dsh@0.2.1-rc.2, pnpm 11.28.5, Caddy 2.11.7) dsh Web UI on 127.0.0.1:3080 behind a Caddy password gate on port 8080, healthcheck /healthz, volume at /data

First login

  1. Open the service's public URL. The browser asks for a username and password.
  2. Sign in with DSH_WEB_USERNAME (default admin) and DSH_WEB_PASSWORD (Railway → deepseek-harness → Variables). You land in the DSH Web UI.
  3. If you did not set DEEPSEEK_API_KEY, open Settings → Models, enter a DeepSeek key, or add another provider.
  4. Choose or create a workspace. The default workspace is /data/home/Documents/deepseek-harness/default-workspace; clone repositories anywhere under /data/home and add them with the folder picker.
  5. Start a session and send a task, for example "Summarize this repository".

Security model: the password protects the page and the hand-out of the Harness's login token; after that, the Harness checks its own signed, HttpOnly session cookie on every API call and WebSocket stream. The agent can run shell commands and spend API credits, so keep the password private and set spending limits at your provider. New sessions use upstream's workspace-write preset: writes limited to the session workspace, approval prompts for anything wider. Read SAFETY.md before giving it access to anything sensitive.

Privacy defaults: OpenTelemetry upload is off (DSH_TELEMETRY_DISABLED=1). Upstream's session-log upload to DeepSeek is on by default when you use the official DeepSeek API; switch it off in Settings → General → Upload Session Log when using the official model API if you prefer.

Custom domain: add it in Railway and list it in DSH_TRUSTED_HOSTS (for example agent.example.com), then redeploy. The Harness refuses API calls from host names it does not trust.

Scaling: the Harness is a single-writer process with file-based state on one volume. Keep one replica and scale vertically; give it more memory if the agent builds large projects.

Pinning and upgrades: the Dockerfile pins DSH_VERSION=0.2.1-rc.2 (npm), PNPM_VERSION, the Node.js and Caddy images. To upgrade, change DSH_VERSION to a newer release from npm and redeploy; the volume keeps settings, credentials and sessions. The project is a developer preview with breaking changes between releases, so read the release notes first.

Why Deploy DSH on Railway?

An agent harness is most useful when it keeps running after you close the lid. Railway gives it an HTTPS address, a persistent volume, automatic restarts and logs without a server to maintain, and you pay for the little RAM and CPU it uses between tasks. Model usage is billed by your provider.

View DSH Coding Agent with Web UI on railway.com

Similar templates

TemplateCategoryDeploymentsDeploys
9Router
Self-hosted AI router for Claude, Gemini, Cursor & OpenAI tools.
AI/ML9,734
flowise-railway
Flowise - low-code LLM apps builder
AI/ML8,512
LibreChat
Open-source ChatGPT clone: multi-model, agents, search, multi-user auth
AI/ML6,762
Hermes Agent by Nous Research
Template for Hermes Agent by Nous Research
AI/ML6,413
OpenClaw
[Homebrew support] The AI that actually does things
AI/ML5,167
Hermes Agent
Self Host Hermes Agent. Autonomous AI agent w/ self-improving, tools & more
AI/ML4,092