rivet-dev/agentos — reverse-engineered prompt

Build me a portable open source runtime for AI agents that I can embed directly into a backend app, instead of spinning up full containers or virtual machines. I want agents to start almost instantly, run in their own isolated VM, and be able to work with files, run shell commands, and execute Node style scripts without touching the host machine directly.

It should support creating and resuming agent sessions, streaming events from a session, saving conversation history automatically, and using a single transcript format across agents. Please include built in support for a few coding agents like Pi, Claude Code, and OpenCode through one unified API. I also want strict deny by default permissions for filesystem, network, processes, environment access, plus CPU and memory limits.

Make it easy to plug in host functions as tools, mount external storage as a filesystem, and support things like webhooks, cron jobs, queues, multiplayer viewing, and durable workflows. It should be self hostable, deployable basically anywhere, and come with a quick start example plus clear docs. Look up current docs online if you need to.