/ballista

Primary LanguageRust

Ballista: Query Your Notes

Introduction

Ever wanted to query your Obsidian notes using ChatGPT? Now you can! This Rust web service lets you do exactly that, by ingesting your Obsidian notes as embeddings into a Qdrant database and then using ChatGPT for prompting.

Usage

Usage assumes that your obsidian-git repo is in a private repo you own.

This service is primarily (currently) deployed through Shuttle. To deploy it, do the following:

  • Run cargo shuttle init --from joshua-mo-143/ballista and follow the instructions
  • Copy Secrets.toml.example to Secrets.toml and fill out the secrets. See the following below for an explanation:
    • OPENAI_KEY: An OpenAI API key for which you have funds on.
    • GITHUB_PERSONAL_ACCESS_TOKEN - A GitHub Personal Access Token. You need at the very least repo:read permissions, since this is required to be able to download your obsidian-git repo if it's private.
    • GITHUB_USERNAME - Your username (or alternatively - someone else's username, if you're using their obsidian-git repo).
    • GITHUB_REPO - The repo you want to ingest. Note that only Markdown files are supported for now - Ballista will ignore anything else.
    • QDRANT_URL - The URL of your Qdrant database. Note that the port needs to be 6334 as qdrant_client utilises the gRPC URL. If you're just trying this out locally, you can leave it blank.
    • QDRANT_API_KEY - Your Qdrant API key. You can leave this blank if you're just running this locally.
  • Run cargo shuttle deploy and watch the magic happen!
  • Visit your deployment URL and try writing a prompt.

Features

  • Downloads your Markdown files from a GitHub repo and ingests it into ChatGPT embeddings, then stores it in Qdrant.
  • Supports Github webhooks at /webhooks/github for self-updating with no input required on your side.
  • Uses an internal queue for resilient updating. If your update fails, simply send the webhook again!

Isn't this just a ChatGPT wrapper?

Yes, but I don't want to pay exorbitant amounts just to be able to query my extremely small knowledge base. If you're interested in using this, I don't think you do either.