PrimeIntellect-ai/prime-agent
Developer ToolsAn open-source coding and research agent with a persistent Python REPL, recursive subagents, and durable harness state for long-running work.
What it is
Prime Agent is an open-source coding and research agent for general and long-running tasks. It is built around two abstractions described in the README: the Recursive Language Model (RLM), which treats context as variables and recursive subagents as function calls inside a persistent Python REPL, and the Continual Harness, which stores supplemental prompts, memories, skill descriptions, and reusable subagent specifications as durable state that can be refined through small updates.
The practical result is that file operations, shell commands, tool use, subagents, and context management all happen programmatically. The agent can spawn child agents with rlm.spawn(...), keep daemon-backed sessions running after the terminal disconnects, and exchange messages between running agents. The harness can be extended with executable Python skills from user configuration; repository skills provide Markdown instructions without automatic code execution.
The project is aimed at developers and researchers who want a persistent coding assistant for long tasks, evaluations, and autonomous workflows, not a one-off chat wrapper.
Quick start
Install the latest stable version with the one-command installer on macOS or Linux:
curl -fsSL https://app.primeintellect.ai/prime-agent/install.sh | sh
On Windows, use PowerShell; the installer downloads the Windows x64 build and verifies its checksum before installation:
irm https://app.primeintellect.ai/prime-agent/install.ps1 | iex
Then start Prime Agent from the repository or directory you want it to work in:
cd /path/to/project
prime-agent
On first launch, run /login to choose a subscription or API-key provider. Prime Agent works in the current directory and can run commands and modify files there. The README recommends using a disposable clone, clean worktree, or another checkpoint you can inspect and restore.
Useful commands include prime-agent agents to browse running and saved sessions, prime-agent attach <agent> to reattach, prime-agent --resume [path|id] to resume, prime-agent status and prime-agent doctor [--fix] to inspect or repair background services, and prime-agent update to update the installation install and getting-started sections.
Constraints
Prime Agent executes model-generated Python and project commands with your user permissions. The worker and kernel processes improve lifecycle isolation and recovery, but the README explicitly says they are not a security sandbox. Run untrusted code or instructions in an external sandbox or restricted environment, and use trusted repositories, instructions, and skills only.
A running session depends on background services; the CLI includes status, doctor, and shutdown commands to manage them. Autonomous mode is bounded by configured turn, token, and time budgets, and a passed quality gate only checks what that gate verifies; reaching a limit does not imply task success.
The project is released under the MIT License per its README.
HotRepo's take
The repository was first seen in HotRepo daily snapshots on 2026-08-07 at 5,643 stars and reached 14,570 stars by 2026-08-12, with daily gains between roughly 1,100 and 2,700 stars over that period. It appeared in the Rust language daily list on 2026-11-17 at rank 5 with 67 new stars, while the overall star count had grown to 22,795. That combination points to continued attention months after the initial surge, though the available snapshots do not include an overall daily rank for this repository. The design emphasis on persistent REPL state, daemon-backed sessions, and a self-refining harness is what distinguishes it in the coding-agent space; the same features mean users should pair it with external sandboxing rather than rely on its process isolation as a security boundary.
Written by AI from the project's README, updated . The project's own docs are the source of truth.
Hacker News
You might also like
- nicobailon/pi-subagentsLets Pi coding sessions hand review, scouting and implementation to focused child agents, including parallel and background runs.
- deepseek-ai/dsh-superagent-prototypeA research prototype of a proactive coding agent that works one GitHub repository on request and replies in the original conversation.
- mvschwarz/openrigCoordinates Claude Code, Codex and other coding agents as a persistent team with roles, shared state and owned work.
- 1jehuang/jcodeA Rust terminal coding agent optimized for running many parallel agents in one repo with low memory and fast startup.
- kirodotdev/KiroCrewA self-hosted persistent workspace where coding agents keep working between sessions, with memory, schedules and approvals.