Quivr setup: document chat with OpenAI and Supabase

Uploaded on May 20, 2023, this tutorial covers an early Quivr setup for storing information and asking questions about documents. The presenter uses Python 3.11.3; the source does not identify a Quivr release version. The installation steps describe that period's application rather than a current setup guide.

The walkthrough clones the repository with Git, installs dependencies, and opens the project in Visual Studio Code. The presenter then configures a secrets file with OpenAI credentials and a Supabase project URL and service key before launching the Streamlit interface on localhost. Although the interface runs on the desktop, this setup relies on external services for model generation and vector storage. It does not demonstrate offline inference or a local LLM.

The interface includes file uploads and URL ingestion, along with controls for chunk size, temperature and token limits. The presenter discusses GPT-3.5 and GPT-4 and proposes using Quivr to retrieve passages or summarize large PDFs through chat. An unresolved API error prevents successful ingestion in the demo and blocks access to some features, so the video does not establish that the complete document question-answering workflow works.

The speaker describes Quivr as open source and makes claims about encryption and privacy. The tutorial does not verify those security claims.