Prime Agent has been rewritten in Rust, with the project reporting faster performance and lower resource use than most coding agent harnesses. The tool, downloaded over 300,000 times since August, orchestrated a swarm of more than 2,000 agents to rewrite itself across 10,000+ Prime Sandboxes, drawing over 200 billion tokens from Prime Inference's GLM-5.3 endpoint. Verification relied on differential testing for TUI, harness, protocol, and feature parity, with separate planner, implementer, reviewer, and verifier agents handling each task.
No score is assigned. Sources and their independence are shown in the citation chain below.