Evolver — GEP-Powered Self-Evolving Engine for AI Agents

Introduction

Evolver is an open-source engine that brings self-evolution to AI agents through Gene Expression Programming (GEP). It maintains auditable Genes, Capsules, and Events that let agents improve their skill libraries autonomously while keeping a full audit trail of every change.

What Evolver Does

• Implements GEP-based evolution for agent skill optimization
• Maintains a versioned library of Genes (atomic skills) and Capsules (skill bundles)
• Records Events that trace every evolution step for auditability
• Supports MCP and A2A protocols for agent interoperability
• Provides a CLI and Node.js API for integration into existing agent systems

Architecture Overview

Evolver models agent capabilities as a population of Genes — small executable skill units that can be composed into Capsules. The evolution loop evaluates Gene fitness against task outcomes, selects high-performing variants, applies crossover and mutation operators, and commits improved Genes back to the library. An event log records every mutation, selection, and fitness evaluation for full traceability.

Self-Hosting & Configuration

• Install via npm as a global CLI tool or project dependency
• Initialize a workspace with seed Genes for your domain
• Configure evolution parameters: population size, mutation rate, fitness function
• Set up persistent storage for the Gene library and event history
• Integrate with existing agents via MCP server mode

Key Features

• Auditable evolution with full event history and Gene versioning
• GEP-based optimization that improves skills over successive cycles
• Capsule system for bundling related Genes into reusable packages
• MCP and A2A protocol support for multi-agent interoperability
• Node.js-based with a fast CLI for interactive evolution runs

Comparison with Similar Tools

• GenericAgent — task-driven skill trees; Evolver uses GEP for population-based optimization
• AutoGPT — action loops; Evolver focuses on skill improvement over time
• DSPy — prompt optimization; Evolver optimizes executable skill code
• LangChain — chain composition; Evolver evolves the chains themselves

FAQ

Q: What is Gene Expression Programming? A: A form of evolutionary computation where programs (Genes) are evolved through selection, crossover, and mutation to improve their fitness on a given task.

Q: Is the evolution auditable? A: Yes. Every mutation, selection, and fitness evaluation is recorded as an Event with full provenance tracking.

Q: Can I seed Evolver with existing skills? A: Yes. Import existing skill definitions as initial Genes to accelerate evolution.

Q: Which agent frameworks integrate with Evolver? A: Any framework supporting MCP or A2A protocols, plus direct Node.js API integration.

Sources

• https://github.com/EvoMap/evolver