A-EVO-Lab/a-evolve

What it solves

A-Evolve provides a universal infrastructure for the autonomous self-improvement of AI agents. It eliminates the need for manual harness engineering by allowing any agent to evolve its own prompts, skills, and memory across various domains using a variety of of evolutionary algorithms.

How it works

The system treats an agent's evolvable state—such as system prompts, skill libraries, and episodic memory—as a standard directory structure on the file system (the "Agent Workspace"). An evolution engine then runs a continuous loop: it lets the agent solve tasks, observes the results, analyzes performance logs to mutate the workspace files, and validates these mutations on holdout tasks before committing them via git-tagging for reproducibility.

Who it’s for

It is designed for AI researchers and developers working on LLM self-optimization, reinforcement learning, and agent architectures who want to test and evolve agents without manually tuning them.

Highlights

• Universal Framework: Supports "Bring Your Own Agent" (BYOA), "Bring Your Own Benchmark" (BYOE), and "Bring Your Own Algorithm" (BYO-Algo).
• Autonomous Post-Training: Capable of post-training large models (up to 550B parameters) with no human in the loop.
• Built-in Algorithms: Includes four reference evolution strategies: adaptive_evolve, adaptive_skill, skillforge, and guided_synth.
• Extensive Benchmark Support: Ready-to-use adapters for SWE-bench, MCP-Atlas, Terminal-Bench 2.0, and SkillsBench.
• Git-Integrated: Every accepted mutation is git-tagged, providing a full audit trail and easy rollback.