A project for versatile AI agents that can run with proprietary models or completely open-source. The meta expert has two agents: a basic Meta Agent, and Jar3d, a more sophisticated and versatile agent.

1. Core Concepts
2. Prerequisites
3. Configuration
   • API Key Configuration
   • Endpoints Configuration
4. Setup for Basic Meta Agent
5. Setup for Jar3d
   • Docker Setup for Jar3d
   • Interacting with Jar3d
6. Roadmap for Jar3d

This project leverages four core concepts:

1. Meta prompting: For more information, refer to the paper on Meta-Prompting (source). Read our notes on Meta-Prompting Overview for a more concise overview.
2. Chain of Reasoning: For Jar3d, we also leverage an adaptation of Chain-of-Reasoning
3. Jar3d uses retrieval augmented generation, which isn't used within the Basic Meta Agent. Read our notes on Overview of Agentic RAG.
4. Jar3d can generate knowledge graphs from web-pages allowing it to produce more comprehensive outputs.

1. Clone this project to your work environment/local directory:

   git clone https://github.com/brainqub3/meta_expert.git

2. You will need Docker and Docker Composed installed to get the project up and running:

   • Docker
   • Docker Compose

3. If you wish to use Hybrid Retrieval, you will need to create a Free Neo4j Aura Account:

   • Neo4j Aura

1. Navigate to the Repository:

   cd /path/to/your-repo/meta_expert

2. Open the config.yaml file:

   nano config/config.yaml

Enter API Keys for your choice of LLM provider:

• Serper API Key: Get it from https://serper.dev/
• OpenAI API Key: Get it from https://openai.com/
• Gemini API Key: Get it from https://ai.google.dev/gemini-api
• Claude API Key: Get it from https://docs.anthropic.com/en/api/getting-started
• Groq API Key: Get it from https://console.groq.com/keys

For Hybrid retrieval, you will require a Claude API key

Set the LLM_SERVER variable to choose your inference provider. Possible values are:

• openai
• mistral
• claude
• gemini (Not currently supported)
• ollama (Not currently supported)
• groq
• vllm (Not currently supported)

Example:

LLM_SERVER: claude

Remember to keep your config.yaml file private as it contains sensitive information.

The basic meta agent is an early iteration of the project. It demonstrates meta prompting rather than being a useful tool for research. It uses a naive approach of scraping the entirety of a web page and feeding that into the context of the meta agent, who either continues the task or delivers a final answer.

python -m agents.meta_agent

Then enter your query.

Jar3d is a more sophisticated agent that uses RAG, Chain-of-Reasoning, and Meta-Prompting to complete long-running research tasks.

Note: Currently, the best results are with Claude 3.5 Sonnet and Llama 3.1 70B. Results with GPT-4 are inconsistent

Try Jar3d with:

• Writing a newsletter - Example
• Writing a literature review
• As a holiday assistant

Jar3d is in active development, and its capabilities are expected to improve with better models. Feedback is greatly appreciated.

Jar3d can be run using Docker for easier setup and deployment.

• Docker
• Docker Compose

1. Clone the repository:

   git clone https://github.com/brainqub3/meta_expert.git
   cd meta_expert

2. Build and start the containers:

   docker-compose up --build

3. Access Jar3d: Once running, access the Jar3d web interface at http://localhost:8000.

You can end your docker session by pressing Ctrl + C or Cmd + C in your terminal and running:

docker-compose down

• The Docker setup includes Jar3d and the NLM-Ingestor service.
• Playwright and its browser dependencies are included for web scraping capabilities.
• Ollama is not included in this Docker setup. If needed, set it up separately and configure in config.yaml.
• Configuration is handled through config.yaml, not environment variables in docker-compose.

For troubleshooting, check the container logs:

docker-compose logs

Refer to the project's GitHub issues for common problems and solutions.

Once you're set up, Jar3d will proceed to introduce itself and ask some questions. The questions are designed to help you refine your requirements. When you feel you have provided all the relevant information to Jar3d, you can end the questioning part of the workflow by typing /end.

• Feedback to Jar3d so that final responses can be iterated on and amended.
• Long-term memory.
• Full Ollama and vLLM integration.
• Integrations to RAG platforms for more intelligent document processing and faster RAG.