What Agent Sandbox does
Agent Sandbox (agentsandbox.co) gives AI agents "a computer and a drive" through one API. An agent sends Python or Bash code to POST /v1/execute; it runs in a sandbox and returns stdout, stderr and any files created. Without a session, each call gets a one-shot sandbox that is destroyed afterwards. With a persistent session, the filesystem and installed packages (pip or apt-get) carry over between calls until you delete the session or it idles out (5 minutes by default). You can upload files, inject them into /workspace before or during a session, pass environment variables, and download what the code produces, such as charts, reports and transformed datasets.
You can call the REST API directly with an API key, use the Python SDK (pip install agentsandbox-sdk), wire it into OpenAI tool calling, or install a plugin for the OpenClaw agent. For coding agents such as Claude Code, Codex, Cursor, Windsurf and Cline, the docs give a one-line prompt that points the agent at a skill file with the full API instructions. Sign-in is with a Google account. The service is operated by Agent Sandbox Inc., under Delaware law per its terms; it says you keep all rights to the code you submit and the output files.
