Build a GPT that actually watches your videos

What it does

StreamRAG is a lightweight Flask server that sits between OpenAI’s GPT store and the VideoDB platform. You upload videos, the server exposes search endpoints via an OpenAPI spec, and a custom GPT queries them to return actual video clips or text summaries instead of making things up.

The interesting bit

The “streaming” part is literal: responses can be compiled video segments served as streaming manifests, not just text RAG. Your GPT becomes a video editor that assembles answers on the fly.

Key highlights

• Upload a video collection once, search across it with natural language
• Returns either text summaries (standard RAG) or actual video clip compilations
• Designed specifically for publishing as a custom GPT in the ChatGPT store
• Free tier: 50 uploads on VideoDB without a credit card
• Includes pre-written OpenAPI spec and GPT prompt templates

Caveats

• Requires hosting your own Flask server (no managed deployment yet; Heroku/Replit support is on the roadmap)
• Tightly coupled to VideoDB’s proprietary platform; not a generic video RAG implementation

Verdict

Worth a look if you already maintain a video library and want to make it queryable inside ChatGPT. Skip it if you need self-hosted vector search or aren’t willing to wire up another SaaS API.