Deploy DeepSeek Harness + Web UI

Deploy and Host DeepSeek Harness Web UI on Railway

Launch a protected, persistent DeepSeek Harness agent workspace in one Railway deployment. Use the full browser UI, keep sessions and workspace files across restarts, and connect DeepSeek or another supported model provider when you are ready—without provisioning a database, Redis, or a second service.

DeepSeek Harness is an open-source AI agent harness with a browser UI, workspaces, plugins, agent presets, and support for multiple model providers. This template runs the official developer-preview package behind a small, auditable Railway gateway that adds protected browser access, dynamic-port support, health checks, and persistent storage.

Developer preview: DeepSeek Harness 0.1 can introduce compatibility-breaking changes. This template pins the tested upstream release and does not claim to be an official DeepSeek deployment.

About Hosting DeepSeek Harness Web UI

The template deploys one GitHub-backed service and one persistent Railway volume mounted at /data. During deployment, you choose a Web UI username and password. Model-provider credentials are intentionally not requested by the Railway template: after signing in, you can configure DeepSeek, OpenAI, Anthropic, OpenRouter, Google, Groq, Moonshot/Kimi, another catalog provider, or a custom compatible API from Settings → Models. You may also choose Configure later.

The service listens on Railway's dynamic PORT, exposes /healthz, restarts on failure, and stores Harness configuration under /data/.dsh plus workspaces under /data/workspace. No database, Redis instance, worker, or object-storage service is added.

What This Template Deploys

Highlights

• Complete browser workflow: this is a usable Web UI, not a shell-only agent runtime.
• Bring your preferred model: configure DeepSeek or another supported provider inside Settings → Models.
• Deploy first, configure later: a provider key is optional during Railway setup.
• Persistent by default: settings, sessions, and workspace files survive restarts and redeployments.
• Small topology: one service and one volume, with no bundled database or cache.
• Security-conscious wrapper: protected login, readiness health check, token redaction, and a fail-closed compatibility patch.

Common Use Cases

• Run browser-based coding and research agents against persistent workspaces.
• Evaluate DeepSeek Harness with your preferred supported model provider.
• Test built-in plugins, agent presets, session history, and workspace tools.
• Maintain an isolated self-hosted Harness environment for development and experimentation.

Dependencies for DeepSeek Harness Web UI Hosting

• A Railway account with usage-based billing enabled.
• A user-chosen Web UI username and a strong password of at least 12 characters; 16 or more is recommended.
• Optional credentials for at least one supported model provider when you are ready to run agents.
• The public deployment wrapper repository, which pins @deepseek-ai/dsh to the tested developer-preview release.

Setup

1. Click Deploy Now.
2. Set HARNESS_ACCESS_USERNAME to 3–64 letters, numbers, dots, underscores, or hyphens.
3. Set HARNESS_ACCESS_PASSWORD to a unique value of 12–256 characters; use 16 or more characters when possible.
4. Deploy and wait for the /healthz check to pass.
5. Open the generated domain and sign in with the credentials you selected.
6. Accept the upstream developer-preview notice, create or select a workspace, and either add a model-provider key or choose Configure later.

Usage

Open Settings → Models to add, edit, or remove provider connections. Provider credentials are written through the authenticated Harness settings API and persist on the Railway volume. Create a session, select a workspace, permission mode, model, and reasoning effort, then submit an agent task from the browser.

The public domain is protected by the template's login gateway. The upstream one-time launch token is exchanged server-side and is redacted from Railway logs.

Persistence

The /data volume stores:

• Harness state and configuration at /data/.dsh.
• Provider settings and installed/configured plugin state managed by Harness.
• Persistent workspaces and agent-created files at /data/workspace.
• Session data maintained by the pinned Harness release.

Do not remove or remount the volume unless you intend to lose this state. Back up important workspaces independently. A single attached Railway volume limits the service to one replica.

Security and Responsible Use

• Treat the Web UI as an administrative agent interface, not a public anonymous application.
• Use a unique strong gateway password and rotate it if it is shared or exposed.
• Add model-provider credentials only through the authenticated Models screen; never place real keys in repository files, template defaults, screenshots, or support messages.
• Grant the narrowest workspace permission mode suitable for a task and review agent actions before using generated output in production.
• Do not place highly sensitive or regulated data in prompts or workspaces unless your provider, Railway plan, and policies explicitly allow it.
• DeepSeek Harness can execute tools and modify workspace files. Follow the upstream SAFETY.md, provider terms, applicable laws, and third-party licenses.

The wrapper carries a minimal version-guarded compatibility patch so the official Models settings UI works on an authenticated non-loopback Railway domain. The image build fails closed if the pinned package or audited source guard changes.

Cost and Resources

Railway charges for the running service, build time, network usage, and persistent volume according to your plan. Model providers bill API usage separately. Actual CPU and memory use depends on plugins, workspace size, concurrent sessions, and agent tasks. Start with one replica and monitor Railway metrics before setting resource limits.

Updates

This template intentionally pins a tested DeepSeek Harness developer-preview version. Template-managed deployments can check for wrapper updates from Railway. Before adopting a newer upstream release, review release notes, the upstream license and SAFETY.md, re-audit the remote-settings compatibility patch, rebuild, and repeat browser, provider, restart, and persistence tests.

Source and maintenance notes: deployment wrapper · official DeepSeek Harness repository

Why Deploy DeepSeek Harness Web UI on Railway?

Railway provides GitHub builds, HTTPS domains, health-gated deployments, environment variables, logs, metrics, restart policies, and persistent volumes in one workflow. This template packages those pieces into a single-service deployment while keeping provider choice inside the product and preserving the upstream developer-preview disclosure.