AI Video Search Engine

Open source

AI Video Search Engine is a TypeScript demo that indexes YouTube video transcripts as vectors so you can ask questions and jump to the matching clip.

avse.vercel.app
AI Video Search Engine homepage screenshot
GitHub stars
630
Last 30 days
+3(+0%)
Last commit
1 yr ago
Repository age
3 years
Self-hosted
Yes

About AI Video Search Engine

AI Video Search Engine, or AVSE, searches inside videos rather than only their titles. The author built it to show that a Google-style index for video content is feasible with technology that is readily available. It extracts the transcript, thumbnail, description and title of a YouTube video, then creates vector embeddings through JigsawStack's multimodal embedding model.

Transcripts are chunked with timestamps and stored in Supabase's PostgreSQL with the pg_vector extension. A question is matched by vector cosine search across the database, a second search finds related chunks inside each result video, and the chunks are mapped back to the timestamped transcript so the relevant clip plays. Summaries come in point form and text, and chat sessions are temporary and not stored.

Hosting it yourself involves Supabase, Hasura on Fly, JigsawStack and Vercel. The repo includes a Hasura migration, a script for indexing whole YouTube channels, and a script to regenerate embeddings for another model. The current model supports over 75 languages.

Key features

  • Transcript extraction from YouTube videos
  • Vector search by question across stored videos
  • Plays the relevant clip using timestamps
  • Video summaries in point form and text
  • Temporary chat sessions about a video
  • Script for indexing whole channels locally
  • Embedding model covering many languages

Good fit for

  • Searching spoken content in YouTube videos
  • Reference design for vector video search
Tags
video-search
rag
vector-search
youtube
supabase
typescript
ai

AI Video Search Engine: questions and answers

What is AI Video Search Engine used for?
AI Video Search Engine is a TypeScript demo that indexes YouTube video transcripts as vectors so you can ask questions and jump to the matching clip. It is a good fit for searching spoken content in YouTube videos and reference design for vector video search.
Is AI Video Search Engine open source?
Yes. AI Video Search Engine is open source. Its source code is on GitHub at yoeven/ai-video-search-engine and is written mainly in TypeScript.
Is AI Video Search Engine free?
Yes. AI Video Search Engine is open source, so the software itself is free to use.
Can I self-host AI Video Search Engine?
Yes. AI Video Search Engine can be self-hosted on your own server or infrastructure.
What are some alternatives to AI Video Search Engine?
Similar open-source tools in the Search category include Orama, Xyne and XERJ. SaaS products in the same category include Amazon OpenSearch Service, Azure AI Search and MomentClip.
Is AI Video Search Engine actively maintained?
The most recent commit to AI Video Search Engine was on 3 March 2025. The project has 630 stars on GitHub, with 3 added in the last 30 days.

Open-source alternatives to AI Video Search Engine

See all

SaaS alternatives to AI Video Search Engine

See all