RAG Agent with memory 💬, that can perform web search without any external API's.
Based on llama3-groq-tool-use model, that can be ran locally, even on your laptop!
Web search: Users can explicitly note what kind of information thay want to find
- Text based
- Images
- Videos
- Memory: Mark has ability to remember chat history, and use it to answer questions
- Clone repo
git clone https://github.com/avtorgenii/minecraft-ai-advisor.git
- Create venv and install dependencies
poetry install
To run Mark you need to:
- Install Ollama
- Pull llama3-groq-tool-use model
ollama run llama3-groq-tool-use:8b
- Make Ollama serve, so model would be accessible from python app
ollama serve
- In project's root directory
chainlit run app/main.py
- Enjoy 💫
During development, when I tried to make model show me video tutorial on how to build Buildcraft quarry, model literally rickrolled me, sending me a link to Never Gonna Give You Up, without performing web search(!!!) 😵.
Shoutout to Meta or groq team for feeding llama a bit too much Reddit threads to make this possible 😤.