Prime Agent is an open-source coding and research agent from Prime Intellect for general and long-running work. It builds on two abstractions. The Recursive Language Model (RLM) treats context as variables and tools, including recursive subagents, as function calls inside a persistent Python REPL. The Continual Harness stores supplemental prompts, memories, skill descriptions, and reusable subagent specifications as durable state that the agent refines through small, evidence-backed updates.
The persistent Python REPL is the model's built-in tool: file operations, shell commands, tool use, subagents, and context management all happen through code. Subagents spawn as real child agents for parallel or background work and return their results programmatically. A refine command reviews the current trajectory and updates the supplemental harness state, never the immutable base system prompt, and recorded snapshots support rollback.
Sessions are daemon-backed, so agents, REPL state, schedules, and subagents keep running when the terminal disconnects and can be reattached later. Running agents can discover one another, exchange messages, and steer each other's work. The agent executes model-generated Python and project commands with the user's permissions, and its worker and kernel processes are not a security sandbox.
Features
- Programmatic control: a persistent Python REPL handles files, shell commands, tools, subagents, and context
- Built-in subagents: spawn child agents for parallel or background work and collect their results in code
- Continual Harness: persists reviewable lessons as prompts, memories, skill descriptions, or subagent specifications, with rollback
- Executable skills: skills are importable Python packages, and a skill creator turns recurring workflows into skills
- Background sessions: daemon-backed agents survive terminal disconnects and can be reattached
- Agent-to-agent messaging: running agents and retained subagents exchange messages and steer each other
- Long-task controls: automatic compaction, persistent goals, heartbeats, schedules, and a bounded autonomous mode with turn, token, and time budgets
- Decision API: an experimental setting that drives a decision model for real-time control tasks
