01
Add a video
Paste a YouTube or Descript link in the app, or harvest YouTube and Vimeo with the browser extension.
How the knowledge layer works
You choose the videos. Harvest makes them useful to your AI.
Every video starts as something you would have to watch, remember, and find again. Harvest turns it into a navigable summary—and then into a private knowledge library built for retrieval and synthesis.
The practical rule: you decide what enters the library; Harvest turns it into something your AI can navigate.
From one link to a connected library
Adding a video starts two connected workflows. The first creates the summary you read beside the player. The second turns that summary and transcript into compact, searchable knowledge.
02
Capture and summarize
Harvest saves the transcript, converts it to timestamped text, and creates the structured outline shown beside the player.
03
Index the useful parts
After summarization, Harvest creates searchable section digests, semantic embeddings, automatic tags, and concepts linked across videos.
04
Ask your own AI
Your connected AI retrieves a few relevant sections first, follows related concepts when useful, and links back to the exact source timestamp.
What gets indexed
The goal is not to dump hours of transcripts into a chat. Harvest stores a compact map that lets an AI start broad, then open deeper evidence only when the question requires it.
Short digests of the ideas taught in each part of the video, ready for precise retrieval.
Tools, people, claims, and recurring ideas connected across the videos in your library.
Semantic embeddings find relevant material even when your question uses different words than the source.
MCP connection
The doorway, not the database
The knowledge layer lives in the private Harvest database. MCP gives an AI permission to search that layer, inspect a video at the right depth, and return evidence with clickable timestamp links.
“Pull the five strongest ideas about AI agents.”
“Which video had that context-engineering explanation?”
“Turn the best material into a workshop outline.”
What is automatic today
A video is ready for knowledge search when its index status is indexed. New, original videos normally reach that state shortly after their summary finishes.
Automatic after you add a video
- Transcript processing
- Structured summarization
- Section and concept extraction
- Embeddings, tags, and concept links
- MCP retrieval with timestamp links
Still deliberately manual
- Choosing what enters your library
- Watching channels or feeds for new uploads
- Approving outside research sources
- Immediate merging of every similar concept
- Retry after every interrupted server process
Build a library your AI can use.
Start with the videos already shaping your work. Harvest makes the important parts searchable, connected, and easy to cite.
Create Account