Concepts

Knowledge bases

A knowledge base (KB) is the central abstraction in Kognita. It is a named, searchable store of text content backed by vector embeddings. All content and search operations belong to a KB.

Data model

A knowledge base has one default vector embedding configuration that determines which embedding model is used and the vector dimension. When you create a KB you can pick a model; the dimension is derived automatically.

What lives in a knowledge base

A knowledge base holds content items: the text you ingest via the API. Kognita processes each item automatically: splitting, embedding, and indexing it so it can be retrieved by search. You interact only with content items; the processing machinery is invisible.

KnowledgeBase
└── Content[]   (text you ingest; Kognita handles the rest)

Vector embedding configuration

Every knowledge base has exactly one embedding configuration. It controls:

• Model: the embedding model used to convert text into vectors. The model is determined by the type field you choose when creating a knowledge base: TEXT_COST_EFFECTIVE, TEXT_HIGH_QUALITY, or TEXT_AND_MEDIA_HIGH_QUALITY.
• Dimension: vector length derived automatically from the embedding model selected by the KB type.

Chunking strategy

Chunking is handled automatically by Kognita; you do not configure it per request. The strategy is fixed by the knowledge base type you chose at creation time. All current KB types use token-based chunking (512 tokens per chunk), which splits text at token boundaries for consistent embedding quality.

To ensure the best retrieval results, ingest content at a reasonable granularity: one article, FAQ entry, or logical document section per content item rather than massive blobs or extremely short fragments.

Limits