home
Prime Agent by Prime Intellect: the independent, evidence-led guide
Prime Agent is Prime Intellect's MIT-licensed, open-source coding and research agent built around a persistent IPython kernel, recursive subagents, and a daemon-backed session runtime. This independent guide collects the 36 canonical routes, 58 audited features, and 95 claim sources so you can verify every statement against v0.7.1.
First-party announcement, documentation, release, or installer.
Which Prime Agent do you mean?
Prime Agent is Prime Intellect's open-source coding/research agent; this site is a separate, unofficial evidence ledger that helps you distinguish harness behavior, benchmark claims, and runtime boundaries before you install anything.
Verified install command
Use the official installer only after you inspect it. The command below is the canonical entry point for this site and is surfaced again in the install guide.
curl -fsSL https://app.primeintellect.ai/prime-agent/install.sh | bashNot a security sandbox
Persistent IPython, worker processes, skills, and extensions all execute with your OS permissions. If you are reviewing untrusted code, isolate it before you launch Prime Agent.
Architecture at a glance
- Persistent IPython keeps state across ordinary turns.
- Daemon workers enable detach/reattach and crash recovery.
- Sessions are append-only JSONL trees, not ephemeral chat logs.
Capability map
Start with install and quickstart if you want a safe first run. Use features and concepts if you want the architecture boundary, or sources if you need claim provenance.
Benchmark caution
Benchmark rows are self-reported unless a reproducible artifact exists. This site labels those claims directly so you can tell launch marketing from independently inspectable evidence.