This template provides a starter for building a custom ChatGPT-style document search using Next.js, OpenAI, and Supabase. It's designed for developers who want to integrate their documentation into a conversational AI interface, enabling users to query information contextually.

How It Works

The system operates in two phases: build time and runtime. At build time, .mdx files are chunked, and embeddings are generated via the OpenAI API. These embeddings are stored in a Supabase PostgreSQL database with the pgvector extension for efficient similarity searches. A checksum mechanism ensures embeddings are only regenerated for changed files. At runtime, user queries are embedded, used to perform a vector similarity search against the stored embeddings, and the most relevant document chunks are injected into an OpenAI GPT-3 prompt for a contextualized response, streamed back to the client.

Quick Start & Requirements

• Install/Run: pnpm dev (after setting up Supabase and environment variables).
• Prerequisites: Docker (for supabase start), OpenAI API Key, Supabase project setup.
• Setup: Requires cloning the repo, copying .env.example to .env, setting OPENAI_KEY, NEXT_PUBLIC_SUPABASE_ANON_KEY, and SUPABASE_SERVICE_ROLE_KEY. Running supabase start is necessary to obtain Supabase keys and initialize the database with pgvector.
• Docs: Blog post

Highlighted Details

• Leverages pgvector for efficient similarity search within PostgreSQL.
• Embeddings are generated at build time to optimize runtime performance.
• Supports streaming responses from OpenAI.
• Includes a mechanism to regenerate embeddings only for modified files.

Maintenance & Community

• Project is part of the supabase-community organization.
• Further details on community channels or roadmap are not explicitly provided in the README.

Licensing & Compatibility

• License: Apache 2.0.
• Compatibility: Permissive license suitable for commercial use and integration with closed-source applications.

Limitations & Caveats

The project relies on external services (OpenAI API, Supabase) which incur costs. The documentation processing is limited to .mdx files by default, requiring conversion for other formats.