Computer Agents

Claude Code in the cloud, with an SDK.

What if Claude Code had an API? And you could build real applications with it? That’s Computer Agents.

We give you Claude-level AI with full execution capabilities—isolated containers, persistent file storage, built-in skills—all accessible through a clean SDK. No infrastructure to manage.

Quick Links

Quick Start

Run your first agent in 3 minutes

SDK Guide

TypeScript SDK for building workflows

API Reference

REST API documentation

Architecture

Infrastructure and security

Three Primitives

Everything you need to build agentic applications:

Agents — The Brain

Configure how your AI thinks. Custom instructions, model selection (Haiku, Sonnet, Opus), reasoning effort, and multi-turn memory.


const agent = await client.agents.create({
  name: 'Code Assistant',
  model: 'claude-sonnet-4-5',
  instructions: 'You are a senior engineer. Write clean, tested code.',
  reasoningEffort: 'medium'
});

Environments — The Computer

Give your agent a real computer. Isolated containers, persistent file storage, installed packages, secrets management.


const env = await client.environments.create({
  name: 'my-project'
});
 
// Configure runtimes and packages
await client.environments.setRuntimes(env.id, ['python', 'nodejs']);
await client.environments.installPackages(env.id, ['flask', 'pytest']);
 
// Files persist across all sessions
await client.files.uploadFile(env.id, {
  path: 'config.json',
  content: '{"debug": true}'
});

Skills — The Tools

Pre-built capabilities that just work. Web search, deep research, image generation, plus any MCP server.


const agent = await client.agents.create({
  name: 'Research Engineer',
  model: 'claude-sonnet-4-5',
  enabledSkills: ['web_search', 'deep_research', 'image_generation']
});
 
// Add custom tools via MCP
await client.environments.update(env.id, {
  mcpServers: [
    { type: 'stdio', name: 'github', command: 'npx', args: ['@anthropic/mcp-server-github'] }
  ]
});

Build Real Workflows

Combine primitives to build multi-step workflows where each step builds on the last:


import { ComputerAgentsClient } from 'computer-agents';
 
const client = new ComputerAgentsClient({
  apiKey: process.env.COMPUTER_AGENTS_API_KEY
});
 
// Create a persistent thread
const thread = await client.threads.create({
  environmentId: 'env_xxx',
  agentId: 'agent_xxx'
});
 
// Step 1: Research
await client.threads.sendMessage(thread.id, {
  content: 'Research best practices for REST API design in 2024',
  onEvent: (e) => console.log(e.type)
});
 
// Step 2: Build (files from step 1 still exist)
await client.threads.sendMessage(thread.id, {
  content: 'Create a Flask API following those practices'
});
 
// Step 3: Test (agent remembers everything)
await client.threads.sendMessage(thread.id, {
  content: 'Write tests and run them'
});

Why Computer Agents?

Raw Claude API	Claude Code	OpenAI Agents SDK	Computer Agents
No execution	CLI only	No infrastructure	Execution + SDK + Managed
No files	Local only	You build it	Persistent workspaces
No tools	Not an API	Patterns only	Skills built-in

Documentation Structure

SDK Guide — TypeScript SDK

Quick Start — Run your first agent
API Resources — All SDK resources
Streaming — Real-time events
Advanced — Production patterns

API Reference — REST API

Authentication — API keys
Threads — Conversations
Environments — Execution contexts
Agents — Agent configuration
Files — File operations

Architecture — Platform details

Infrastructure — Cloud components
Security — Security model
Scalability — Scaling behavior

Get Started


npm install computer-agents


import { ComputerAgentsClient } from 'computer-agents';
 
const client = new ComputerAgentsClient({
  apiKey: process.env.COMPUTER_AGENTS_API_KEY
});
 
// This is Claude Code, but as an API
const result = await client.run(
  'Create a Flask API and run it',
  {
    environmentId: 'env_xxx',
    onEvent: (event) => console.log(event.type)
  }
);
 
console.log(result.content);

Get your API key →