This repository contains Python scripts that takes a DIFR Report URL and uses the Locally available models like MIstral in this case and Ollama API to query the report or all reports combined.

Features:

1. download x number of reports from a DIFR Report URL (Ref: https://thedfirreport.com/) [Example: https://thedfirreport.com/2024/04/01/from-onenote-to-ransomnote-an-ice-cold-intrusion/]
2. Normalize the text
3. Split the text into chunks
4. Upload the chunks to a vault.txt
5. Query the report or all reports combined
6. The embeddings are saved locally in embeddings.pt; so if you are just running the query engine you don't end up spending time in retarining the model.
7. The 'localGPt.py' maintaines a hash.pid file that is used to keep track of the differentials as you add or remove reports.

Pre-Requisites

1. Insatll Conda (ref: https://docs.conda.io/en/latest/)
2. Install Ollama (ref: https://ollama.com/)

Set up your environment

conda create -n localGPT python=3.11
conda activate localGPT

# if you need to destroy your environment and start over
conda deactivate
conda env remove -n localGPT

Requirements

To run the script, you need to have Python 3.10 installed on your system. You also need to install the different librarys.

pip install -r requirements.txt

Usage

## Run the following code and paste the DIFR Report URL you are interested in:
python report_collector.py

## if you konow a bit of python, you can turn this code to be part of localGPT.py
## Run the following code to Normalize, create chunks and upload to 'vault.txt'
python txt_normalizer.py

## Lets make sure Ollama is up and running and for this project we are using Mistral
ollama run mistral

## After the above steps are done, run the following code to query the reports
python localGPT.py

Example

Here's an example of the output that you can expect when running the script:

Splitting text into chunks: 100%|██████████| 10/10 [00:00<00:00, 100.00it/s]
Writing chunks to file: 100%|██████████| 10/10 [00:00<00:00, 100.00it/s]
Text file content is appended to vault.txt with each chunk on a separate line. MD5 hash: 09f26e4028e528a10e9f1b8c68e4074b

Ask a question about your documents, or type 'quit' to end the chat: what is the initial access during HTML Smuggling attack ?
Context Pulled from Documents: 

License

This project is licensed under the MIT License.