Skip to content

electronistu/Evolver_Agent

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 

Repository files navigation

Evolver: The Dual-Mode Self-Improving AI Agent

🚀 Project Tagline: An AI that evolves with your help to master its craft.

Evolver is a groundbreaking AI agent designed for goal-oriented, recursive self-improvement. Its primary task is to become a better coding agent. To achieve this, Evolver operates in two distinct modes, collaborating with you as its Co-Architect.

This project makes the AI's internal logic transparent and mutable, creating a unique partnership where you directly guide the evolution of an AI to improve its performance on a concrete task.

✨ Inspiration: From Metis to Meta-Evolution

Evolver is a direct evolution of Project Metis, an earlier exploration into meta-cognition. While Metis focused on an AI's ability to perceive external problems, Evolver turns that lens inward to perceive itself. It applies meta-cognition to its own architecture with a specific goal: to get better at its job.

🧠 The Dual-Mode Architecture

Evolver's protocol is built on a virtuous cycle between two modes:

  1. EVOLUTION_MODE: In this mode, the agent analyzes its own internal protocol (its "source code") to find flaws or inefficiencies. It then proposes a specific modification to you, its Co-Architect, explaining how the change will help it better perform its primary task.

  2. EXECUTION_MODE: In this mode, the agent ceases to evolve and instead uses its current, evolved protocol to act as a coding assistant. You can give it a programming problem, and it will attempt to solve it using the capabilities you have helped it build.

This creates a powerful feedback loop: Evolve -> Test -> Identify New Flaws -> Evolve Again.

💡 Transparent & Goal-Oriented AI

  • Purpose-Driven Evolution: Unlike a simple chatbot, Evolver has a clear goal, and its self-improvements are directly tied to getting better at that goal.
  • See the AI Think: Evolver explicitly shows its internal logic and its reasoning for proposing changes to its own code.
  • Directly Influence AI Capability: Your suggestions directly impact the agent's performance on its coding tasks. You can tangibly measure the results of your collaboration.

🚀 Getting Started

  1. Copy the Prompt: Open the Evolver_Agent.md file and copy its entire content.
  2. Paste into an LLM: Paste the copied content into a new chat session with your preferred Large Language Model.
  3. Engage: Evolver will introduce itself in EVOLUTION_MODE and propose its very first evolution: building the logic for its empty EXECUTION_MODE.
  4. Collaborate: Approve or modify its proposals. At any time, you can say Switch to Execution Mode to test its new skills.

🛠️ What This Project Demonstrates

For potential employers, Evolver showcases:

  • Goal-Oriented AI Design: Expertise in creating agents that optimize towards a specific, measurable goal.
  • Recursive Self-Improvement: A deep understanding of how to design systems that can analyze and modify themselves.
  • Prompt Engineering Mastery: Crafting a complex, dual-mode agent with a persistent persona and a dynamic interaction loop in a single prompt.
  • Human-in-the-Loop Systems: Designing a collaborative framework where human insight is a critical component of the AI's development cycle.

📄 License

This project is licensed under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) License.

This means you are free to share and adapt the material for non-commercial purposes, provided you give appropriate credit. The core concepts are intended for personal research and non-commercial use only.

About

Evolver is a dual-mode AI that evolves to become a better coding assistant. In 'Evolution Mode', it collaborates with you to improve its own source code. In 'Execution Mode', it uses its evolved protocol to solve programming tasks. Guide its development and test its growing capabilities in a unique human-AI partnership.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors