Knowledge | Agno

Knowledge is a class that manages knowledge bases for AI agents. It handles adding content, searching, and deleting content from the knowledge base.

Parameter	Type	Default	Description
`name`	`Optional[str]`	`None`	Name of the knowledge base
`description`	`Optional[str]`	`None`	Description of the knowledge base
`vector_db`	`Optional[VectorDb]`	`None`	Vector database for storing embeddings
`content_db`	`Optional[BaseDb or AsyncBaseDb]`	`None`	Preferred constructor argument for the content metadata database.
`contents_db`	`Optional[BaseDb or AsyncBaseDb]`	`None`	Supported read/write alias of `content_db`. If both arguments are supplied, they must be the same object.
`page_store`	`Optional[FileSystem]`	`None`	Enables coordinated published pages; requires a PostgreSQL-backed filesystem and PgVector in the same logical database.
`page_search`	`Optional[PageSearchConfig]`	`None`	Configuration for published-page retrieval.
`max_results`	`int`	`10`	Maximum number of results to return from searches
`query_transformer`	`Optional[QueryTransformer]`	`None`	Rewrites the query before it reaches the vector database, for example with `HyDE`. Rerankers still score against the original query. See Query Transformation.
`reranker`	`Optional[Reranker]`	`None`	Reranks search results. `Knowledge` fetches a wider candidate pool sized by the reranker, reranks it, and returns the top `max_results`. See Reranking.
`isolate_vector_search`	`bool`	`False`	When `True`, adds `linked_to` metadata during insert and filters by it during search. Enables isolation when multiple Knowledge instances share the same vector database. See Isolate Vector Search.
`content_sources`	`Optional[List[BaseStorageConfig]]`	`None`	Cloud storage providers (S3, GCS, SharePoint, GitHub, Azure Blob) for loading remote content. See Cloud Storage Sources.
`readers`	`Optional[Dict[str, Reader]]`	`None`	Dictionary of custom readers for processing content
`max_embedding_retries`	`int`	`0`	Extra attempts when embedding fails during ingestion. Each retry re-embeds the whole document. Authentication and oversized-chunk failures are not retried. See content status.
`embedding_retry_backoff`	`float`	`1.0`	Seconds before the first retry. The wait doubles on each later retry.

The constructor is keyword-only. contents_db remains the canonical dataclass field for serialization and replacement; the alias is supported, not deprecated. Call setup() or asetup() before using a configured page store. See published-page methods and result types.