Evolver is an open-source self-evolution engine for AI agents built around what its authors call the Genome Evolution Protocol — a structured way to turn agent run logs into auditable, versioned prompt and tool updates instead of ad hoc tinkering. Teams deploy an agent, collect traces from production or evaluation runs, and then feed those traces back to Evolver, which proposes focused changes to system prompts, tool definitions, and workflow logic as reviewable evolution assets.
The model is deliberately protocol-bound: every change the engine suggests is captured as a discrete genome diff with its motivating logs attached, so engineers can read, approve, reject, or roll back updates the same way they review code. This is a sharp contrast to the usual 'edit the system prompt and pray' loop most agent teams fall into, and it gives agent behavior a proper change history instead of leaving tuning locked inside individual developers' heads.
Evolver is GPL-3.0 licensed and ships as a JavaScript toolkit that wraps around existing agent stacks rather than replacing them. It is best suited to teams that already run an agent in production and want repeatable, inspectable improvement cycles — regulated environments, internal copilots, and any setting where agent changes need an audit trail.
