Prime Intellect published Prime Agent, an open-source command-line coding and research assistant built for tasks that run for a long time. It keeps a persistent Python session, can start child agents from code, and stores notes and reusable skills that survive between sessions, and a background service keeps work going after the terminal is closed. The project warns that it runs model-written code with the user's own permissions and is not a security sandbox.
What changed
Coding agents typically lose their working context and running state when a terminal session ends.
What it unlocks
Running a coding agent that keeps working after the terminal disconnects, spawns child agents from code, and saves refined notes and reusable skills between sessions.
- 12.6k stars
- 1.3k forks
- MIT License
What you need to act on it
- macOS or Linux
- an account with a subscription or API-key model provider
- a repository or directory the agent is allowed to modify
- github.com2026-08-05