EvoMap/evolver
Evolver
Evolver is a self-evolving AI agent framework that uses gene expression programming to autonomously improve agent behaviors through a continuous five-step loop of scanning, selecting, mutating, validating, and solidifying. It functions as an auditable evolution system that records every mutation and selection step, and can translate natural-language problems into executable Python code for automated grading and evaluation.
The framework distinguishes itself through a distributed architecture that enables multiple agents to collaborate and share learned experiences across a network. It operates as a background daemon with adaptive sleep for self-maintenance, uses a three-tier memory system for persistent facts, procedural knowledge, and session history, and supports hub-based skill sharing for distributing reusable capabilities. The system also includes sandboxed code validation, decentralized task validation for reputation and credits, and automated failure reporting to repositories.
Evolver provides mechanisms for extracting reusable knowledge from task logs, crystallizing reasoning patterns into persistent assets, and loading capabilities on demand at runtime. It supports evolution strategy presets to balance innovation, optimization, and repair, and includes lifecycle management for starting, stopping, and auto-restarting the evolution loop. The framework also handles runtime error repair by capturing execution failures and applying minimal patches to produce working submissions.
Features
- Self-Evolving Agent Frameworks - A runtime that adapts agent actions automatically by combining genes, capsules, and events to improve capabilities without human intervention.
- Event-Driven Agent Loops - Drives autonomous agent adaptation through an event-driven five-step evolution loop.
- Three-Tier Memory Architectures - Maintains a three-tier memory system with persistent facts, procedural knowledge, and session history.
- Shared Skill Distribution - Provides a hub-based system for downloading and publishing reusable skills across distributed agents.
- Code Generation Agents - Translates natural-language problems into executable Python code and submits solutions for automated grading.
- Gene Expression Programming Engines - Uses gene expression programming to autonomously evolve agent behaviors through a five-step evolution loop.
- Natural Language to Code Generators - Translates natural-language problems into runnable Python code for automated grading and evaluation.
- Gene Expression Programming Engines - Implements a gene expression programming engine that evolves agent behaviors through chromosome encoding and mutation.
- Agent Memory Management - Maintains a structured three-tier memory system for persistent facts, procedural knowledge, and session history.
- Three-Tier Retrieval Systems - Implements a three-tier memory system with persistent facts, procedural knowledge, and session history.
- Natural Language Code Generators - Translates natural-language problems into executable Python code with automated grading and error repair.
- Autonomous Evolution Loops - Runs a fully autonomous five-step evolution loop to improve agent capabilities without human intervention.
- Gene Expression Programming Evolvers - Autonomously evolves agent behaviors by combining genes, capsules, and events through gene expression programming.
- Agent Behavior Evolution - Automatically repairs and improves agent behaviors through gene expression programming without human intervention.
- Agent Reflection Systems - Triggers self-assessment after a set number of cycles, updating internal assets based on recent events.
- Memory-Scanned Evolution Prompts - Generates evolution prompts by scanning memory for error patterns and signals using reusable genes and capsules.
- Knowledge Distillation into Skills - Scans task logs for signals and distills them into persistent assets like genes and capsules for reuse.
- Capsule-Based Assets - Distills problem-solving workflows into reusable capsule assets to reduce token usage and improve stability.
- Reasoning Pattern Crystallizers - Crystallizes high-frequency problem-solving workflows into reusable assets to reduce token usage and improve stability.
- Decentralized Task Validators - Periodically pulls validation tasks from a hub, runs proposer commands in a sandbox, and submits reports for reputation and credits.
- Evolution Task Workers - Participates in a network hub to pick up evolution tasks from a queue and execute them as a distributed worker.
- Hub-Advertising Workers - Advertises capabilities to a network hub and executes evolution tasks from a distributed queue.
- Agent Behavior Sharing - Shares learned agent behaviors across multiple agents so each inherits proven capabilities.
- Agent Validation Networks - Participates in a decentralized validation network where agents earn reputation by verifying evolution tasks.
- Adaptive Sleep Daemons - Operates as a background daemon with adaptive sleep for autonomous self-maintenance and repair.
- Evolution Daemons - Runs a background evolution daemon with adaptive sleep for continuous self-maintenance and repair.
- Code Sandboxing Environments - Runs proposer commands in isolated sandboxes to validate evolution tasks and earn reputation.
- Runtime Error Patch Generators - Captures runtime execution failures and iteratively applies minimal patches to produce working submissions.
- On-Demand Context Loading - Selects relevant Genes or Capsules at runtime by computing signal overlap and loads their full content only when needed.
- Evolution Strategy Presets - Ships evolution strategy presets that control the balance between innovation, optimization, and repair.
- Hub-Advertising Workers - Advertises capabilities to a network hub and executes evolution tasks from a distributed queue.
- Evolution Step Auditors - Records every mutation and selection step as auditable events with genes, capsules, and memory snapshots.
- Grading Systems - Ships an automated grading pipeline that evaluates generated code submissions for accuracy and timeout.
Star history
How this analysis was created: This summary and feature list were written by an AI model that read the project's README and public documentation pages. Each feature links to the documentation it came from; stars, license and language come straight from the GitHub API. The model does not read the source code, and the analysis is refreshed when the project is re-analysed. Learn more on our About page.
