abhiinnovates/evolver

Evolver is a tool for teams who manage AI agents and want to improve them over time in a structured, trackable way. Instead of manually tweaking an agent's prompts and hoping for the best, this engine analyzes the agent's runtime logs, spots error patterns, and generates a carefully guided prompt to help the agent evolve and improve. The core idea is turning ad hoc prompt changes into repeatable, auditable "evolution assets" that can be reused and shared. The tool works by scanning a project's memory directory for logs and historical signals about how the agent is performing. Based on those signals, it selects a matching "Gene" or "Capsule" (reusable templates for behavior) and outputs a strict, protocol-bound prompt. That prompt is meant to guide the next step in the agent's development. Importantly, Evolver does not edit source code on its own or run arbitrary shell commands, it produces a prompt that a developer or a host runtime then acts on. This is built for teams maintaining AI agents at scale, especially in environments that need an audit trail for compliance or review. For example, if a customer support agent keeps failing at a certain task, a team member could run Evolver to generate a targeted repair prompt, apply it, and record the whole event for traceability. It is not meant for one-off scripts or projects that need free-form creative changes. It also requires Git, using it for rollback and safety calculations. A few notable things: the project offers different strategy presets, so teams can bias the engine toward shipping new features, hardening stability, or focusing purely on emergency repairs. It can run in a continuous loop or be triggered by platform hooks. Optionally, teams can connect it to the EvoMap network to share reusable skills, pick up evolution tasks from a shared worker pool, and participate as a decentralized validator. The core functionality works fully offline.