Two ways to use it
1. Run the agent inside the sandbox
Use Treadstone when the agent itself should live in the isolated environment and keep working there over time.
2. Use the sandbox as a tool
Use Treadstone when your agent runs elsewhere, but needs an isolated environment it can create, inspect, and drive on demand.
In practice, that means an agent can:
create and manage sandboxes over the CLI, REST API, or Python SDK
operate what is running inside them through shell, file, browser, HTTP/WebSocket, and MCP
hand the same browser session to a human when review or takeover matters
Treadstone is an agent-native sandbox platform for longer-running AI work. You can run agents inside isolated sandboxes, or treat sandboxes as tools that agents call on demand. Each sandbox is built around an all-in-one runtime: code execution, shell, file system, browser-facing surfaces, MCP, and long-running state.
Treadstone is open source, self-hostable, and built on Kubernetes.
2. Use the sandbox as a tool Use Treadstone when your agent runs elsewhere, but needs an isolated environment it can create, inspect, and drive on demand.
In practice, that means an agent can:
create and manage sandboxes over the CLI, REST API, or Python SDK operate what is running inside them through shell, file, browser, HTTP/WebSocket, and MCP hand the same browser session to a human when review or takeover matters