A knowledge store is a persistent, queryable store of your videos and images plus the understanding the platform derives from them. It contains three layers:

Spatiotemporal context: entities, moments, facts, and summaries extracted from your videos and images
Typed ontology: a schema that organizes those elements and their relationships
Embeddings: vector representations that make the content semantically retrievable

Together these layers form the foundation for corpus-level reasoning: the platform reasons across the full collection, not just individual clips.

Ingestion configuration

The ingestion configuration controls what the platform extracts from your videos and images. Set it when you create a knowledge store.

For the available approaches and guidance on choosing one, see the Configure ingestion guide.

Knowledge store items

A knowledge store item is an asset added to a knowledge store for indexing. Items are processed asynchronously.

Add asset → queued → pending → processing → ready
                                          → failed

Status	Meaning	Action
`queued`	Waiting to be processed	Poll and wait
`pending`	About to start processing	Poll and wait
`processing`	Indexing in progress	Poll and wait
`ready`	Item indexed and queryable	Proceed to query
`failed`	Indexing failed	Check error, retry with a new item

All items must reach the ready status before queries return meaningful results. Each query targets a single knowledge store. You cannot query across stores.

Jupyter notebook

API reference

POST /knowledge-stores - create a knowledge store
GET /knowledge-stores/{id} - retrieve a knowledge store
POST /knowledge-stores/{id}/items - add an item
GET /knowledge-stores/{id}/items/{item_id} - check item status