Welcome to MatrixGPT, a project striving to implement the innovative concepts presented in the research paper by Stanford University and Google Research.
This repo only contains a portion of those concepts:
- A minimalist grass park environment for AI agents to navigate with ease
- A pair of OpenAI-integrated agents equipped with memory and conversational capabilities
- Although the agents possess the functionality to generate reflections, this feature is currently disabled in this specific instance.
👁 View the project proposal 👁
The MatrixGPT runs locally on your machine. To run:
- Update the
env.jsonfile with your OpenAI API key - In the root of the project's directory run
npm installto install all dependencies - Then run
npm run devin the root directory. This will spawn the front-end and create 2 agents named Edward and Rosalie. Open your browser tohttp://localhost:3000 - To view the agent's conversation open your browsers debug console
Implementing all aspects of the research paper is an overwhelming task for a single developer. If you find this project promising, I encourage you to contribute by selecting one of the following items to work on:
- The ability for agents to generate and react to plans
- Agents should be able to move on their own according to their plans or based on "free will"
- Agents need to know how to save memories based on conversations they've had
- An architecture that handles more 2+ agents conversing at a time
- Add places to the map such as housing, shops, and restaurants.
- UI enhancements so that we can interact with the agents
- Personality traits that can evolve over time
- Emotions that affect the tone of an agent's voice
- Need a timing system (day/night cycles)
- Implement agent actions (e.g dig a hole)