Skip to main content
The Knowledge API organizes project content for retrieval. Document metadata, content chunks, and embedding vectors represent different parts of the indexed content.

​
Documents and chunks

A document record stores metadata about content, such as its title, source, and retrieval attributes. File chunks divide source content into pieces that can be indexed and returned as search matches. Project source files remain part of the Projects API. Updating a document’s metadata is separate from editing its source file or replacing its indexed chunks. An embedding is a numeric vector associated with a content chunk. Generate a vector through the AI Gateway API or another supported provider, then store it for retrieval. Vector search compares a query vector with indexed content. The selected vector dimension must match the relevant stored embeddings. Text-based knowledge lookup searches metadata in project knowledge files. For example, a support agent can look up a policy file by its frontmatter, or retrieve indexed chunks by vector similarity. These retrieval paths use different inputs and return different kinds of matches. An uploaded file is not automatically part of either source lookup or a vector index.

​
Content versions

Branches, environments, and releases can identify different versions of project content. A search or chunk read applies to the version selected by its endpoint and parameters. Branch chunk operations support changes to the index. Environment and release reads expose content for those versions. Each reference page defines the available selectors, pagination, and update preconditions.

​
Get started

Add project knowledge shows a source-file lookup. The knowledge ingestion guide covers indexed content.

​
API references