Evolver is an open-source self-evolution engine for AI agents: it turns the scattered lessons of past agent runs into auditable, reusable assets โ Genes and Capsules โ that make agents measurably better over time. Published by EvoMap under the GPL-3.0 license, it has collected more than 8,600 stars on GitHub since its early-2026 release.
Most agent improvement today is ad-hoc prompt tweaking that nobody can review or roll back. Evolver formalizes it with the GEP (Gene Expression Programming-style) protocol: the CLI scans a project’s logs and memory, selects a Gene or Capsule from its asset pool, and emits a structured evolution prompt that a host runtime (Claude Code, Codex, Cursor, OpenCode, OpenClaw and others) can consume. Every cycle is written as an EvolutionEvent for audit, review mode lets a human approve changes before they land, and an optional connection to the EvoMap network adds shared skills, a worker pool and evolution leaderboards. The engine works fully offline on its own.
Evolver is free and open source under the GPL-3.0 license (earlier MIT/GPL releases keep their original terms; the project has announced future releases will move toward source-available). The optional EvoMap network registration is free.
A public changelog and frequent tagged releases โ more than 140 in the project’s first months โ make the evolution of the engine itself easy to audit, very much in keeping with the project’s governance theme.
It is for agent builders and teams who run coding or task agents daily and want their agents’ hard-won experience captured as governed, reusable assets instead of evaporating into chat logs.