primeintellect-ai/prime-agent
A coding agent that keeps its lessons after the chat window closes and edits its own notes
Prime Agent is an open-source agent for coding and long-running autonomous work, built around a persistent Python REPL (a live, line-by-line code execution environment) that treats context as variables and sub-agent calls as function calls. A '/refine' command reviews the current work trajectory and applies small, evidence-backed updates to durable state such as extra prompts, memories, and skill descriptions, while leaving the base system prompt untouched. Because it runs as a daemon (a background process), sessions keep going after the terminal disconnects and can be reattached later.
What it does
- A persistent Python REPL is the built-in tool: file operations, shell commands, tool use, sub-agents, and context management all happen through code
- rlm(...) spawns real child agents for parallel or background work and returns their results programmatically
- /refine applies only small, evidence-backed updates to the harness (durable supplemental state) and records history so changes can be rolled back
- Skills are importable Python packages, and a built-in skill creator turns recurring workflows into reusable skills
- Persistent goals (/goal), heartbeats, schedules, and a bounded autonomous mode (/autonomous) keep long tasks progressing across turns and sessions
Why it matters
It lets long research evaluations or repetitive coding workflows continue without a human staying attached, and lets the agent carry forward what it learned into future sessions. The README explicitly warns this is not a security sandbox, since the agent executes model-generated Python and commands with the user's own permissions.
Terms in this repo
- RLM (Recursive Language Model) · an approach treating context as variables and sub-agent calls as function calls
- Continual Harness · durable state that stores supplemental prompts, memories, and skill descriptions across sessions
- REPL · an interactive environment that executes code line by line and shows results immediately
- daemon · a background process that keeps running after the terminal disconnects
- skill · a reusable Python package capturing a recurring workflow
Repository description (English)
A self-improving RLM agent for coding workflows and long-running autonomous tasks.
Open on GitHubCoverage
Trending repos
- tt-a1i/archifyAn AI agent skill that turns a described codebase into a verified, shareable architecture diagram inside chat
- bookorbit/bookorbitA self-hosted book server that syncs your reading progress across web, Kobo, and KOReader
- ConardLi/garden-skillsA skill pack that hands AI coding agents ready-made expertise for design, video, images, and search
- browser-use/browser-useA tool that lets AI click buttons and type into websites just like a human, to finish tasks for you
- K-Dense-AI/scientific-agent-skills163 how-to manuals that turn AI coding assistants into science research helpers
- DietrichGebert/ponytailA ruleset that teaches AI coding agents to say no to writing code
- liustack/modlensA plugin that lets text-only AI models like DeepSeek 'see' pasted images by turning them into structured text data
- marin-community/marinMarin is an open-source platform that publishes the entire process of building large language models, not just the final model
Latest from METAL LAB
- Pika, WAN 3.0에 얼굴 2개·배경 1개 따로 지정해 시대극 제작
- Prime Intellect publishes technical report on Prime Agent, ARC-AGI-3 score jumps from 30% to 95.5%
- Perplexity unveils agent memory system "Brain"
- Qwen's New Model Qwen3.8-Flash-Next Runs Locally on 75GB of Memory
- OpenAI Discloses Full Account of Hugging Face Breach