Execution infrastructure for AI agents

SandrPod turns the machines you already pay for into on-demand sandboxes for agent code. It is a control plane, a worker, and an in-sandbox execution engine — three processes you can run on one laptop, one server, or a fleet across eight cloud providers.

The agent does not know the difference. Neither does its SDK.

Run anywhere you own : Eight cloud providers, plain Docker, or a bare machine with no Docker at all (sandrpod-agent). No inbound ports on workers: they dial out.

Speak any SDK : Native REST, Python, TypeScript, LangChain / deepagents, and the unmodified E2B SDK as a drop-in. Change one environment variable, not your code.

Stay in control : Self-hosted data, opt-in permission gate, decision audit, per-sandbox egress allow-lists, and credential masking the sandbox cannot see through.

Sixty seconds

curl -O https://raw.githubusercontent.com/sandrpod/sandrpod/main/docker/docker-compose.yml
docker compose up -d

That pulls three published images and starts a control plane on localhost:8080 plus one Docker worker that dials back over a reverse tunnel. No Go toolchain, no build, no clone.

pip install sandrpod-cli
sandrpod-cli create demo --provider local
sandrpod-cli execute demo "echo hello from a sandbox you own"

Where each reader should start

You want to	Read
Run a sandbox in the next five minutes	Quickstart
Keep using your E2B code	E2B drop-in
Drive it from a script or an agent	CLI · REST API
Understand the moving parts	Architecture
Put it in front of real traffic	Self-host in production
Run it as a paid service	docs.sandrpod.com is the engine; the console is a commercial build

What this is not

SandrPod is not a hosted product you sign up for. The engine is Apache 2.0 and runs entirely on infrastructure you control. There is a hosted console at cloud.sandrpod.com built on the same engine, but nothing in this documentation requires it.