Prime Agent is an open-source TypeScript-based coding and research agent designed for general and long-running autonomous tasks. The project is built around two core abstractions: the Recursive Language Model (RLM), which treats context as variables and tools as function calls within a persistent REPL environment, and the Continual Harness, which stores supplemental prompts, memories, skill descriptions, and subagent specifications as durable state that can be refined through evidence-backed updates.
The agent combines a persistent Python control environment with durable harness state, enabling useful working context and reusable operating patterns to persist beyond individual chat sessions. Everything in Prime Agent is programmatic, with persistent IPython serving as the built-in model tool for file operations, shell commands, tool use, and subagent management. Subagents are a first-class feature, spawned via rlm(...) calls to enable parallel or background work with results returned programmatically. The harness can improve itself through a /refine command that reviews trajectories and applies small, evidence-backed updates to supplemental state while preserving the immutable base system prompt and supporting rollback through recorded snapshots.
Skills in Prime Agent are executable Python packages, and the built-in skill creator allows recurring workflows to be converted into project or personal skills. Sessions can run in the background through daemon-backed agents that continue executing when the terminal disconnects and can be reattached later. Running agents communicate directly with one another without routing all interactions through the user, and long-running tasks maintain progress through automatic compaction, persistent goals, heartbeats, schedules, autonomous mode, and retained subagents across turns and terminal sessions.
The repository has grown steadily, with stargazers increasing from 5789 to 5886 and forks from 463 to 469 since early August 2026. Issue and pull request response latency shows a median of 0.5 hours and mean of 3.3 hours across 49 tracked items, indicating active maintenance. The most frequently addressed issue labels are bug (3 occurrences), documentation (2), and enhancement (1). Primary contributors tracked by GitGenius include sethkarten with 9 events, ayushnangia with 5 events, and trungtly with 3 events. The project shares overlapping contributors with related repositories including anthropics/claude-code, modernizr/modernizr, and openhands/openhands.
Prime Agent is fully open source under the MIT License and includes comprehensive documentation covering quickstart, usage and CLI reference, long-running and background agents, the RLM programming model, JSON and RPC modes for headless automation, skills management, provider setup, architecture overview, and development guidance. The agent is built on top of the pi project and includes CI and binary build workflows. Installation is available for macOS and Linux, with the installer verifying SHA-256 checksums and preparing the IPython runtime. Users must review changes and use only trusted repositories, instructions, skills, and extensions, as Prime Agent executes model-generated Python and project commands with user permissions in a non-sandboxed environment.
