Search knowledge in Elasticsearch with vector, keyword and hybrid search

Agno's new Elasticsearch vector database, from agno.vectordb.elasticsearch, stores the chunks of an Agno Knowledge in an Elasticsearch index. Teams that already run Elasticsearch for search and logs can keep agent knowledge there too.

from agno.knowledge.knowledge import Knowledge
from agno.vectordb.elasticsearch import Elasticsearch
from agno.vectordb.search import SearchType
 
knowledge = Knowledge(
    vector_db=Elasticsearch(
        index_name="recipes",
        url="http://localhost:9200",
        search_type=SearchType.hybrid,
    ),
)

Vector search is the default. It runs kNN over a dense_vector field and applies your filters before the search. Keyword search runs multi_match over each chunk's content and name. Hybrid search combines the two, and hybrid_strategy picks how:

• HybridStrategy.boost, the default, adds 0.7 of the vector score to 0.3 of the keyword score. It works on every Elasticsearch license.
• HybridStrategy.rrf uses reciprocal rank fusion. It needs a Platinum, Enterprise or trial license. On a cluster without one, Agno logs a warning and falls back to boost.

The Elasticsearch vector database connects with a URL, a list of node URLs, or an Elastic Cloud ID, and authenticates with an API key or basic auth. It verifies TLS certificates by default. It also supports metadata filters, rerankers, bulk inserts and upserts, and per-user isolation through a user_id field. Every read and write has an async version on the native async client.

Pin the elasticsearch client to your cluster's major version. An 8.x cluster rejects a 9.x client.

See the cookbook, and learn more about the Elasticsearch vector database and per-user isolation in the documentation.