Exact-RAG is a powerful multimodal model designed for Retrieval-Augmented Generation (RAG). It seamlessly integrates text, visual and audio information, allowing for enhanced content understanding and generation. This repository contains the source code, and example scripts to facilitate the usage and exploration of Exact-RAG.

• Python >= 3.10
• Poetry (Install)

• To use OpenAI models you need a valid key at this link
• To use local LLM models you need Ollama. Here the instructions to install

You should download a LLM model usin ollama. By default the orca2 model is set in settings.toml, to download just run:

ollama run orca2

• To use Elasticsearch you should have a running cluster or you can use a test deployment using the docker-es.sh script.

Clone the repository:

poetry install # -E audio -E image

• audio extra will install openai-whisper for speech-to-text

• image extra will install transformers and pillow for image captioning

First step is to modify the settings.toml file.

Then starting the web server just running:

poetry run python exact_rag/main.py

NOTE: The first start up could required some time to download the selected models, expecially for image captioning

UI Demo is build upon streamlit and it is made just to demo purposes. If you want to run locally to quick try eXact-RAG features, be sure to have install the packege with dev dependencies, and then:

poetry run streamlit run frontend/ui.py

You can find some examples of usage in the examples folder

Chat with images example:

Chat with PDF and Tables:

To run the tests:

• be sure to have install the dev dependecies with:

  poetry install --with-dev # -E audio -E image

• then run:

  poetry run pytest tests/

We welcome contributions! If you'd like to contribute to Exact-RAG, please follow our contribution guidelines.

• Fork the repo.
• Clone the repo from your codebase
• Choose your favorite editor and open the folder.
• Create a new branch: git checkout -b
• Make changes, commit them and push it back up to github using git push origin .
• Open pull request on GitHub.

This project is licensed under the MIT License.