We are Vijit and Jack working on Runtime Use (
https://docs.runtimeuse.com/): an open source runtime that lets you run agents (Claude, Open AI, custom, etc) in any sandbox.
It provides a simple runtime + client layer for controlling agents in a sandbox:
- start an agent server inside a sandbox (npx -y runtimeuse@latest)
- send prompts, stream outputs, and exchange files over websocket
- cancel or manage long-running tasks
By default it runs an OpenAI Agents SDK agent with shell + web search tools. You can switch to Claude code agent with --agent=claude.
A Python client connects to the sandbox and lets you interact with the agent programmatically. Docs and quickstart available at the linked website.
We built this while working on https://getlark.ai/ where agents run in sandbox environments to generate end-to-end tests. Getting the infrastructure right took significant effort, so we extracted and open-sourced the runtime layer.
We’re excited to hear any feedback, thoughts, questions, and concerns in the comments!
It provides a simple runtime + client layer for controlling agents in a sandbox: - start an agent server inside a sandbox (npx -y runtimeuse@latest) - send prompts, stream outputs, and exchange files over websocket - cancel or manage long-running tasks
By default it runs an OpenAI Agents SDK agent with shell + web search tools. You can switch to Claude code agent with --agent=claude.
A Python client connects to the sandbox and lets you interact with the agent programmatically. Docs and quickstart available at the linked website.
We built this while working on https://getlark.ai/ where agents run in sandbox environments to generate end-to-end tests. Getting the infrastructure right took significant effort, so we extracted and open-sourced the runtime layer.
We’re excited to hear any feedback, thoughts, questions, and concerns in the comments!