Skip to Content
Overview

Welcome to Testbase

Testbase is a unified SDK for building multi-agent systems with seamless integration between LLM-based and computer-use agents. Create agents that think (LLM) and agents that execute (computer), with runtime abstraction that lets you switch between local and cloud execution effortlessly.

This documentation provides comprehensive guides for getting started, understanding the architecture, and building production-ready agent workflows.

What you can do

  • Create computer agents quickly. Spin up Codex-powered agents that can read files, write code, run tests, and execute terminal commands—all within a safe workspace.
  • Compose multi-agent workflows. Combine LLM agents (for planning and review) with computer agents (for execution) to build sophisticated automation pipelines.
  • Run anywhere. Switch between local execution (fast, for development) and cloud execution (isolated, with billing) by simply changing the runtime configuration.
  • Integrate MCP servers. Connect your agents to external tools and services using the Model Context Protocol—same configuration works for both agent types.

Core concepts

Two agent types:

  • 'llm' - OpenAI API for reasoning, planning, and review
  • 'computer' - Codex SDK for code execution and file operations

Two runtimes:

  • LocalRuntime - Execute on your local machine (fast iteration)
  • CloudRuntime - Execute on GCE VM (isolated, billable, production)

Key features:

  • Automatic session continuity across multiple tasks
  • Unified MCP server configuration
  • Manual workflow composition (you control the flow)
  • Professional billing system with usage tracking
  1. Get Started - Install the SDK and run your first computer agent in 30 seconds
  2. Architecture - Understand how agents, runtimes, and MCP servers work together
  3. Agents SDK - Dive into agent configuration, session management, and workflow patterns
  4. Cloud Platform - Learn about cloud execution, billing, and production deployment

Quick example

import { Agent, run, LocalRuntime } from 'computer-agents';
 
// Create a computer agent that runs locally
const agent = new Agent({
  name: 'Developer',
  agentType: 'computer',
  runtime: new LocalRuntime(),
  workspace: './my-project',
  instructions: 'You are a software developer.'
});
 
// Execute a task
const result = await run(agent, 'Create a Python calculator');
console.log(result.finalOutput);
 
// Continue the conversation (automatic session continuity!)
await run(agent, 'Add error handling');

Documentation structure

  • Foundations - Core concepts, architecture, and quickstart guides
  • Agents SDK - Detailed agent configuration, MCP integration, and patterns
  • Cloud Platform - Production deployment, billing, and API reference
  • Workflows - Advanced patterns like tool integration and automation
  • Reference - Configuration reference, glossary, and further reading

Staying current

This documentation lives alongside the Testbase web application and is updated as new capabilities are added. The canonical source of truth is the main repository:

  • SDK Documentation: computer-agents/CLAUDE.md
  • Quick Start: computer-agents/QUICK_START.md
  • Cloud Infrastructure: computer-agents/packages/cloud-infrastructure/README.md
  • Billing System: computer-agents/packages/cloud-infrastructure/BILLING.md

If you notice gaps or want to propose improvements, open a pull request or file an issue on GitHub.

Last updated on
Agents SDK Overview