# Agent Sandbox > A secure runtime API for AI agents to execute Python and shell commands in sandboxed environments with file handling. Agent Sandbox provides a trusted runtime environment for AI agents to execute untrusted code safely. It offers a simple API that allows agents to run Python and shell commands inside secure, isolated workspaces with automatic dependency installation and file management capabilities. The platform enables developers to build AI agents that can actually execute tasks by providing them with a personal remote computer and drive. Whether building code interpreters, agent workflows, artifact pipelines, or codegen evaluations, Agent Sandbox handles the secure execution layer. - **Secure Code Execution** allows AI agents to run Python and shell commands inside safe, sandboxed workspaces with configurable memory and timeout settings. - **Automatic Dependency Management** installs libraries and system tools automatically and reproducibly for each run by simply listing what your agent needs. - **File Operations** enable uploading data for agents to process and downloading everything they create including charts, reports, and transformed datasets. - **Session Persistence** maintains workspace state across multiple executions, allowing agents to write files in one execution and read them in subsequent ones. - **Framework Agnostic Integration** works as a single tool inside any agent framework including Claude Code, Codex, Cursor, Windsurf, and Cline with minimal setup. - **Python SDK** provides a clean interface with `pip install agentsandbox-sdk` for easy integration with async support and comprehensive error handling. - **REST API** offers direct HTTP endpoints for code execution, session management, and file operations with Bearer token authentication. To get started, sign up to receive an API key and $10 in free credits. Install the Python SDK or use the REST API directly. Execute code by calling `client.run("print(2 + 3)")` and retrieve results including stdout, stderr, and generated files. Use sessions for persistent workspaces where state carries across executions. ## Features - Secure Python and shell code execution - Automatic dependency installation - File upload and download - Persistent session workspaces - Environment variable injection - Python SDK with async support - REST API access - Multi-agent framework integration - Usage-based billing - Real-time execution tracking ## Integrations Claude Code, Codex, Cursor, Windsurf, Cline, OpenAI, OpenClaw Bot ## Platforms WEB, API, DEVELOPER_SDK ## Pricing Paid ## Links - Website: https://www.agentsandbox.co - Documentation: https://www.agentsandbox.co/docs - EveryDev.ai: https://www.everydev.ai/tools/agent-sandbox