Prime Intellect has open-sourced Prime Agent, a self-improving coding harness that replaces fixed tool schemas and context compaction with a persistent Python REPL and a rewritable "Continual Harness." The system treats sub-agent delegation as function calls within this environment, allowing models to manage their own prompts, skills, and memory.

  • Prime Agent uses the Recursive Language Model (RLM) abstraction, where sub-agents are launched as child sessions via `rlm("sub-task")` without blocking the parent loop.
  • The Continual Harness formalizes state as H = (ρ, G, K, M), enabling agents to create, read, update, and delete their own trajectory components.
  • A `/refine` command allows the agent to apply minimal edits to its own prompt or skills based on past performance, with rollback capabilities by ID.
  • On ARC-AGI-3, Prime Agent with Opus 5 reports 95.5% RHAE Best@1, surpassing the human expert baseline of 95.4%.
  • The tool supports various backends including subscription logins, API keys, and self-hosted models like GLM-5.2, installing via a single command on Linux or macOS.

Prime Agent is designed for long-horizon tasks such as overnight refactors, spec-driven builds, and kernel optimization, offering lower token usage than native harnesses by running functions over data rather than reading data through tools.