Chippr-AGI is an open-source event-driven ECS framework that uses AI models to automate task creation and prioritization. This system is designed to run solo or as a swarm of containers. It combines the power of LLM with actor-critic reinforcement learning to optimize the order and allocation of tasks for a given objective.
5/9 - https://docs.chipprbots.com/ is live, needs work
- Clone the repository:
git clone https://github.com/chippr-robotics/chippr-agi.git - Install the dependencies:
yarn - Update your API keys docker-compose configuration located in
./docker/docker-compose.yml - Update the OBJECTIVE in
./examples/simple_demo.js - Start redis (you can use docker-compose for this)
- Start the application:
yarn demo - Monitor the vector DB at
http://localhost:8001
Add systems and components to meet your needs then yarn start
First, pull the Chippr-AGI Docker image from Docker Hub:
docker pull chipprbots/chippr-agi:latest
To run the Chippr-AGI container, you'll need to set up environment variables for your OpenAI API key and Redis credentials. Create a .env file by copying .env.example
Replace CHIPPRAGI_LANGUAGE_MODEL_API_KEY, CHIPPRAGI_VECTORDB_HOST, CHIPPRAGI_VECTORDB_PORT, and with your actual values.
Now, run the container with the following command:
docker run -d --name chippr-agi --env-file .env chipprbots/chippr-agi:latest
This will start the Chippr-AGI container in detached mode and load the environment variables from the .env file.
Add the value for CHIPPRAGI_LANGUAGE_MODEL_API_KEY to the docker-compose.yml located in ./docker
docker-compose up
Chippr-AGI uses a combination of GPT-4 for generating task descriptions and actor-critic reinforcement learning to prioritize the tasks based on their estimated rewards. The framework is built using Node.js and Redis to store embeddings for quick query and update.
Tasks are generated based on the current context and objective, which are passed into a customizable prompt template. The prompts are stored in a JSON file and can be easily edited to fit specific needs. Once a task is generated, its dependencies are added to the task list and prioritized based on their estimated reward.
After a task is completed, the system checks if any new tasks need to be generated based on the success of the previous task. The process is repeated until all tasks are completed and the objective is achieved.
graph TD
A(User) -- objective --> B(EntityCreationSystem)
C -- tasks, parent --> B
B -- parent --> C(TaskParentSystem)
B -- entity --> D(System Selection System)
D -- systemselected --> F(task creator)
D -- systemselected --> G(image creator)
D -- systemselected --> H(internet Search)
D -- systemselected --> I(task expander)
F -- tasks --> B
G -- taskCompleted --> J(The Judge)
H -- taskCompleted --> J
I -- taskCompleted --> J
J --> K[task completed]
K -- yes --> L(TaskCompletionSystem_TBD)
K -- no --> D
In this flowchart:
- The objective is provided to the system by a user (see examples/simple-demo)
- Tasks are stored as entities with associated components.
- An event is emitted to addSystemSelection to the task, which is handled by the System Selection system.
- We add a component to each entity mapping it to its parent objective
- An event is emitted to addSystemSelected the task, which is handled by the system selection system.
- The system selection system evaulates which loaded systems can best complete the task
- An event is emitted identifying which system will process the task
- The task is executed based on the relevant components.
- The result of the task is stored, and the task is marked as done.
- The system checks if the objective is complete.
- If the objective is not complete, an event is emitted to generate new tasks, which is handled by the Task Generation system.
The ChipprAGI class emits an event using the EventEmitter. The EventEmitter distributes the event to all registered systems. Each system handles the event if it's relevant to that system. The system performs its specific action based on the event and updates the relevant components. This diagram shows a high-level overview of how events are propagated through the ChipprAGI system and how systems handle and react to events.
graph TD
A(ChipprAGI) -->|Emit event| B(EventEmitter)
B -->|Distribute event| C(System 1)
B -->|Distribute event| D(System 2)
B -->|Distribute event| E(System N)
C -->|Handle event| F(Perform System 1 specific action)
D -->|Handle event| G(Perform System 2 specific action)
E -->|Handle event| H(Perform System N specific action)
F --> I(Update relevant components)
G --> J(Update relevant components)
H --> K(Update relevant components)
We welcome contributions from the community. If you'd like to contribute to Chippr-AGI, please fork the repository and submit a pull request. We recommend discussing your ideas with the community in the issues section before starting any major work.
This project is licensed under the APACHE-2.0 License. See the LICENSE file for more details.