Sandbox Agent

Sandbox Agent lets developers control coding agents from their own applications through a shared API. It's for teams building agent-powered software that needs remote sessions and isolated code execution without a separate integration for each agent.

• Shared agent interface: Control Claude Code, Codex, OpenCode, Amp, or Pi through the same HTTP API and standardized session events.
• Session control: Create sessions, send messages, and handle permission requests from your application while agents run inside a sandbox.
• Activity records: Stream tool calls, file edits, and permission events to storage you choose, such as Postgres or Rivet, for later replay and auditing.
• Built-in debugger: Inspect sessions and event payloads to investigate agent behavior without building a debugging interface.

The server can run locally, on your own infrastructure, in Docker, or inside E2B, Daytona, and Vercel Sandboxes. Local execution is available through the TypeScript SDK; choosing a hosted sandbox provider places execution in that provider's environment.

Node and browser applications can use the TypeScript SDK, and applications in other languages can connect through HTTP. Sandbox Agent is open source under Apache 2.0.