Bhodi The IA powered doc explorer

Bhodi is a chatbot framework to work with vectorized documentation.

---

Set env

uv venv -p 3.10
source .venv/bin/activate

mkdir models # store your models here
mkdir data_to_vector # store the data to vectorize here

⚠️ IMPORTANT!

Check config.py and configure your models, embedders, and tokenizers. I suggest putting your models in the models directory.

Hint: You can download models from Huggingface.

Install Packages

Packages

• Install pyproject packages

uv sync

Llama.cpp

• For Nvidia CUDA:

export CMAKE_ARGS="-DGGML_CUDA=on"
uv pip install llama-cpp-python

Llama.cpp Installation: Check official repo of llama.cpp to see how to install lama.cpp for your hardware.

Build bhodi

• build our cli tool to index data in vectorstore and your caller for the chatbot
• execute this in root project directory

uv build
uv pip install dis dist/bhodi_doc_analyzer-0.1.0-py3-none-any.whl

• to use bhodi-index:

bhodi-index path/to/your/data # Support file or directory (only trully tested with PDF)

• to use bhodi as a chat bot

bhodi

---

---

FEATURES

🌟 ACTUAL FEATURES / IMPLEMENTED

• Basic chatbot TUI
• Vectorize/index several type of files
• Chat logs
• Easy to use
• Chat Memory with RAG

⚙️ NEEDED FEATURES / NEED TO BE IMPLEMENTED (PRIORITY)

• Copy MD blocks of code generated by the chat (maybe a widget in textual)
• Better indexing of files = more consistent and accurate RAG

💡 COULD BE USEFUL / NEED TO BE IMPLEMENTED (NOT PRIORITY)

• Embeding models using API, like google-gemini or OPENAI