Kindle GPT enables AI-powered search and chat on Kindle highlights, allowing users to query their reading notes and generate answers based on them. It targets Kindle users who want to leverage their highlights for deeper understanding and knowledge retrieval, offering a local-first solution for managing and interacting with this personal data.

How It Works

The application parses an exported Kindle notebook (.html file) to extract highlights. It then generates embeddings for each highlight using OpenAI's text-embedding-ada-002 model, combining chapter/section names with the highlighted text for relevance. These embeddings are stored locally, and users can query them via a search bar that calculates cosine similarity to find the most relevant passages. The top results are used to construct a prompt for GPT-3.5-turbo to generate an answer.

Quick Start & Requirements

• Install dependencies: npm i
• Run app: npm run dev
• Requires an OpenAI API key.
• Recommended for desktop use only.

Highlighted Details

• Utilizes OpenAI Embeddings (text-embedding-ada-002) for highlight embedding.
• Supports .csv export and import of embedded data.
• Stores all data locally, without a database.
• Queries leverage cosine similarity for relevance.
• Generates answers using GPT-3.5-turbo.

Maintenance & Community

The project is maintained by mckaywrigley. Contact is available via Twitter.

Licensing & Compatibility

Code is 100% open source. No specific license is mentioned in the README, implying a permissive license by default, but this should be verified.

Limitations & Caveats

The README recommends desktop-only usage, suggesting potential issues with mobile or browser compatibility. The maximum token limit for returned results is approximately 2k tokens.