22.1K ·GitHub
What is this?
Prime Agent is an open-source coding and research agent for general and long-running work. It is built on two abstractions: the Recursive Language Model (RLM) programming model and a Continual Harness that keeps prompts, memories and skills as durable, refinable state.
Prime Agent is an open-source coding and research agent for general and long-running work. It is built on two abstractions: the Recursive Language Model (RLM) programming model and a Continual Harness that keeps prompts, memories and skills as durable, refinable state.What does the RLM part actually change about how it works?Context becomes variables and tools become function calls inside a persistent Python REPL — file operations, shell commands and subagents all run through code. Calling rlm(...) spawns real child agents for parallel or background work and returns their results programmatically.
Context becomes variables and tools become function calls inside a persistent Python REPL — file operations, shell commands and subagents all run through code. Calling rlm(...) spawns real child agents for parallel or background work and returns their results programmatically.How do I install and start it?On macOS or Linux, run the installer: it downloads the latest stable release, verifies the SHA-256 checksum, installs the prime-agent command and can prepare the Python runtime. Then start prime-agent from the repository or directory you want it to work in.
On macOS or Linux, run the installer: it downloads the latest stable release, verifies the SHA-256 checksum, installs the prime-agent command and can prepare the Python runtime. Then start prime-agent from the repository or directory you want it to work in.Do I need my own API keys or a subscription?You authenticate with /login on first launch, which lets you pick either a subscription or an API-key provider — whichever you already have.
You authenticate with /login on first launch, which lets you pick either a subscription or an API-key provider — whichever you already have.Is it safe to point at my main repository?It executes model-generated Python and project commands with your full user permissions — worker and kernel processes add lifecycle isolation and recovery, but this is not a security sandbox. The README advises a disposable clone or clean worktree you can inspect and restore.
It executes model-generated Python and project commands with your full user permissions — worker and kernel processes add lifecycle isolation and recovery, but this is not a security sandbox. The README advises a disposable clone or clean worktree you can inspect and restore.Is it just for coding, or other work too?It is a coding and research agent, but the README stresses long-running work — especially evaluations in research — with compaction, persistent goals, heartbeats and schedules to keep tasks moving across turns.
It is a coding and research agent, but the README stresses long-running work — especially evaluations in research — with compaction, persistent goals, heartbeats and schedules to keep tasks moving across turns.Is it open source?Yes — fully open source under the MIT License.
Yes — fully open source under the MIT License.What can the agent improve about itself?The /refine command reviews the current trajectory and can apply small, evidence-backed updates to supplemental harness state — prompts, memories, skill descriptions and reusable subagent specs. It never rewrites the immutable base system prompt, and recorded snapshots support rollback.
The /refine command reviews the current trajectory and can apply small, evidence-backed updates to supplemental harness state — prompts, memories, skill descriptions and reusable subagent specs. It never rewrites the immutable base system prompt, and recorded snapshots support rollback.What happens if my terminal disconnects mid-task?Sessions are daemon-backed: the Python REPL state, schedules and subagents keep running after detach and can be reattached later. /goal keeps an objective active across turns until completed or paused, and heartbeats or schedules re-enter a session periodically or at a set time.
Sessions are daemon-backed: the Python REPL state, schedules and subagents keep running after detach and can be reattached later. /goal keeps an objective active across turns until completed or paused, and heartbeats or schedules re-enter a session periodically or at a set time.Can I leave it running on its own?Yes — /autonomous continues within configured turn, token and time budgets and can run user-defined quality gates. Mind the caveat: a passed gate only checks what that gate verifies, and reaching a limit does not imply the task succeeded.
Yes — /autonomous continues within configured turn, token and time budgets and can run user-defined quality gates. Mind the caveat: a passed gate only checks what that gate verifies, and reaching a limit does not imply the task succeeded.Can I package my own recurring workflows into it?Skills are importable Python packages, and a built-in skill creator turns recurring workflows into project or personal skills. For headless automation and integrations there are also JSON mode and RPC mode.
Skills are importable Python packages, and a built-in skill creator turns recurring workflows into project or personal skills. For headless automation and integrations there are also JSON mode and RPC mode.What are similar projects?ruflo, deer-flow, OpenHarness, Orbital, Step-Code, Raven, pstack-claude, open-pstack
ruflo, deer-flow, OpenHarness, Orbital, Step-Code, Raven, pstack-claude, open-pstackWhich topics is it in?self-improving-agents, subagents, harness, autonomous-agents, agent-memory, monorepo, tui, agentic-ai, multi-agent, coding-agent, agent-skills, rust
self-improving-agents, subagents, harness, autonomous-agents, agent-memory, monorepo, tui, agentic-ai, multi-agent, coding-agent, agent-skills, rust