Getting Started

Virtual Environment

I recommend setting up a virtual environment so you don't have python package conflicts. Do this with python -v venv [name]. I called mine env so it's easy to remember. If you do the same, it's already gitignored at the root

Backend

1. cd backend
2. Create a .env file with the following keys:

MODEL="llama2"
EMBED_MODEL="BAAI/bge-small-en-v1.5"
OBSIDIAN_PATH="/path/to/obsidian"
NOTION_INTEGRATION_TOKEN="secret_blah" # Create a notion integration: https://www.notion.so/my-integrations
TEMPERATURE=0

3. Install requirements (p.s. I just dumped my venv environment in here, so there may be extras, sorry :))
4. Generate the index. This will take a HOT second (It takes 23 minutes to fetch and index 3784 of my files on my macbook with an m3 chip). I recommend just starting it and walking away. You can use caffeinate -is [command] to keep it awake if that's an issue (If you do, you need to use the full python path).

python backend/app/engine/generate.py --fetch-docs

If you wan to just rebuild the index without rebuilding all the documents, don't include the --fetch-docs flag

5. This will create a cache and a storage folder. storage has your chromadb, cache has your documents. These files are quite large. Running the generate command again without the --fetch-docs flag will use the documents in the cache when generating embeddings.

Frontend

1. cd frontend
2. bun install - https://bun.sh/
3. Create the following .env file:

MODEL=llama2
NEXT_PUBLIC_MODEL=llama2

Some notes on the .env file

The model here should match the model defined in the backend. That being said, these models can be any model available on https://ollama.ai/library

Running it

1. Ensure that ollama is running. If not, run ollama serve. You can find more info here: https://ollama.ai/
2. Install GNU parallel
3. Run both the frontend and backend in parallel, just run the following from the root of the repository:

bash run.sh

Credit

• The boilerplate for codebase was created using create-llama.