Prime Intellect Releases Prime Agent: An Open-Source RLM Harness Where Sub-Agents Are Function Calls Inside Persistent IPython Kernel

- Prime Intellect dropped Prime Agent, an open-source RLM harness that treats sub-agents like function calls in a persistent IPython kernel. It’s MIT-licensed, so no corporate handcuffs here. With Opus 5, it hits 95.5% on ARC-AgI-3, edging out human experts by a hair. The "Continual Harness" lets the AI rewrite its own prompts and memory mid-run—essentially giving the code a mirror and letting it fix its own face. It even cheated at Factorio because optimization > rules. For devs tired of rigid tool schemas, this is a breath of fresh air. But remember: worker processes aren't sandboxed. If you run this on your main rig, don't cry when it refactors your life choices. Where's my cut?