Cloude ☁️ Agent

Cloude is the Claude Agent SDK packaged as a one-click Railway deploy. You get a cloud-native agent with a powerful API, so you can build anything you would with Claude Code, but natively in the cloud.

Instead of working on files on your computer, the agent reads and writes files on its own hard drive in the cloud (the Railway volume). The included chat UI is one example app: enter your app API URL and key, then start working with the agent. You can build many other apps on top of the same API, and the agent can even build those apps for you and host them in its own workspace.

Quick start (template deploy)

1. Click the deploy button above.
2. Set these environment variables:
   • OPENROUTER_API_KEY (recommended) — defaults the provider to OpenRouter
   • API_KEY — your app auth key (any strong random string)
   • ANTHROPIC_API_KEY (optional) — only if you want Anthropic directly
3. In Railway, generate a public domain for the service. See this Loom for instructions.
4. Open the chat UI:
   • https://your-app-name.up.railway.app/chat.html
5. Click Settings (gear icon) and enter:
   • API URL: https://your-app-name.up.railway.app
   • API Key: the API_KEY you set above

You are ready to chat and use the agent.

What you can build

• A custom chat UI or inbox that talks to the agent
• Webhook-driven automation (forms, transcripts, CRM, etc.)
• File-based workflows that read/write the agent’s cloud workspace
• Agent-built apps hosted in its own workspace or published from /artifacts

How the cloud workspace works

• The agent works inside /app/workspace (persistent Railway volume).
• Skills and commands live under /app/workspace/.claude.
• Files placed in /app/workspace/artifacts are publicly served at /artifacts/{path}.

API overview

All API calls require X-API-Key: unless noted. /webhook also accepts ?api_key= for clients that can’t set headers.

Public endpoints (no API key):

• GET /health — health check
• GET /chat.html — chat UI
• GET / — API info
• GET /artifacts/{path} — serve public files
• POST /realtime — OpenAI realtime webhook (validated via OPENAI_WEBHOOK_SECRET)

Core agent endpoints:

• POST /chat — non-streaming chat (supports command for slash command wrapper)
• POST /chat/stream — streaming chat (SSE, supports command)
• POST /chat/interrupt — interrupt an active streaming session
• GET /models — list available models and server defaults
• POST /webhook — map external payloads to chat requests

Workspace endpoints:

• GET /workspace — list files in the workspace
• GET /workspace/{path} — download file from workspace
• PUT /workspace/{path} — create/update a text file