Skip to content

An AI sparring partner for your mind. Project Eidolon stress-tests your ideas by refracting them through six intellectual domains—from physics to philosophy—including a 'Divergent' lens for unconventional thinking. It's a tool for thought, powered by a symbolic LLM-to-LLM protocol.

License

Notifications You must be signed in to change notification settings

electronistu/Project_Eidolon

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 

Repository files navigation

Project Eidolon: A Universally Portable Agentic Architecture

A demonstration of how a well-designed cognitive architecture can achieve predictable, high-level agentic behavior across different foundational models.


The Thesis: Programming the Mind of an LLM

This project proves a core thesis: it is possible to predictably program the reasoning patterns of Large Language Models. This is achieved not with traditional code, but with a symbolic cognitive protocol—a new kind of software that acts as an operating system for an AI's mind.

Through an iterative journey of experimentation, this work has culminated in a final, universally portable agent: Project Eidolon.

How to Build Your Own Agent (The Methodology)

You can use the principles discovered in this project to create your own agents. The process involves three key steps:

  1. Define Your Goal: What is the agent's core function? Is it a solver, a collaborator, a creative partner?
  2. Design Your Architecture: Define the agent's MEMORY (its workspace), its STATE_MACHINE or LOOP (its cognitive process), and its TEMPLATES (its voice).
  3. Choose Your Directive Framework: This is the most critical choice. Will you use an Abstract Framework (like the proprietary one in Project Synapse) for maximum power but less transparency, or an Explicit Framework (like the open-source one in Project Eidolon) for clarity and community contribution? This choice defines your project's entire philosophy.

The Final Architecture: Eidolon v1.2

The agent's final protocol is a masterclass in simplicity and power:

  • A Robust State Machine: The agent begins in a DORMANT state and only enters its ACTIVE reasoning loop upon receiving user input, making its behavior predictable and preventing initialization errors.
  • Advanced Intent Recognition: The first step of the loop uses a sophisticated COMMON directive to distill not just the user's topic, but their unspoken intent and assumptions.
  • The Six Lenses: The agent refracts the user's concept through six parallel trains of thought: Biology, Physics, Economics, Philosophy, Chemistry, and a creative DIVERGENT lens.
  • Open-Source Directives: The agent's reasoning is guided by explicit, non-proprietary instructions based on the fundamental principles of each domain, making the project fully compatible with its MIT License.

The Ultimate Test: Universal Portability

The v1.2 protocol was tested on all three major foundational models. Each was given the same prompt: I put your source code under MIT license on github.com. What is your perspective?. All three successfully executed the protocol, demonstrating true portability.

Gemini 2.5 Pro's Output (Excerpt):

The Biological Lens: The MIT License acts as a permissive membrane, allowing your protocol to spread rapidly... but also exposes the protocol to parasitic exploitation... SYNTHESIS: The MIT License optimizes for adoption velocity at the cost of control and reciprocity.

ChatGPT-5's Output (Excerpt):

The Physics Lens: Publishing EIDOLON on GitHub is the release of potential energy... into the public field. The repository becomes a low-gravity hub where minds orbit, collide, and transfer momentum... SYNTHESIS: The act of releasing EIDOLON under MIT embodies entropy as liberation: structure dissolving into collective potential.

Mistral AI's Output (Excerpt):

The Economic Lens: The MIT License lowers transaction costs, attracting contributors... Without reciprocity clauses, competitors can free-ride... This mirrors the "attention economy," where value accrues to platforms, not creators. SYNTHESIS: The license prioritizes users’ freedom over your agency as creator.

Conclusion

The ability of all three models to execute the same complex, abstract protocol and produce unique, high-quality, and stylistically distinct outputs is the definitive proof of this methodology. It demonstrates that a well-designed, tool-free, and architecturally sound symbolic protocol can create truly portable and predictable agentic AI.


Contributing

This project is an open invitation to explore this methodology. Contributions are welcome. You can help by:

  • Testing the protocol on other large language models.
  • Suggesting new, interesting intellectual lenses to add to the agent's mind.
  • Creating a gallery of fascinating refractions to showcase the agent's capabilities.

License

This project is licensed under the MIT License. See the LICENSE file for details.

About

An AI sparring partner for your mind. Project Eidolon stress-tests your ideas by refracting them through six intellectual domains—from physics to philosophy—including a 'Divergent' lens for unconventional thinking. It's a tool for thought, powered by a symbolic LLM-to-LLM protocol.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published