RAGIT (rag-it) is a git-like software that turns your local files into a knowledge-base. The main goal of this project is to make knowledge-bases easy-to-create and easy-to-share.
rag init;
rag add --all;
rag build;
rag query "What makes ragit special?";
RAGIT is very different from the other RAG frameworks.
- It adds a title and summary to every chunks. The summaries make AIs very easy to rerank chunks.
- It uses tfidf scores instead of vector searches. It first asks an AI to generate keywords from a query, then runs tfidf search with the keywords.
- It supports markdown files with images.
- It supports multi-turn queries (experimental).
- You can clone/push knowledge-bases, like git.
Ragit is primarily supported on Linux (x64) and Mac (aarch64). It goes through a full test process before each release, on Linux and Mac. It is primarily developed on Linux and Mac.
Ragit works on Windows, but it's not perfect.
Other than those 3 platforms, I haven't tested ragit on any platform.
- Build
- Chunks
- Configuration
- Contribution
- Evaluation
- Multi Turn
- Pipeline
- Prompt Engineering
- Quick Guide
cargo install ragit;
rag clone https://ragit.baehyunsol.com/sample/ragit;
cd ragit;
# The default model is groq's llama.
# If you have groq api key, you can use the model.
export GROQ_API_KEY=YOUR_API_KEY;
# If you want to use another model, you can change the model like this.
rag config --set model gpt-4o;
export OPENAI_API_KEY=YOUR_API_KEY;
rag query "How do I contribute to ragit?";