Update: PgVector now supports similarity_threshold!

Stop returning low-quality matches. Set a quality floor on your vector search results so your agents only work with context that's actually relevant.

Getting started is simple

Pass similarity_threshold when initializing PgVector. Anything below the threshold gets filtered out automatically.

from agno.vectordb.pgvector import PgVector
 
vector_db = PgVector(
    table_name="vectors",
    db_url=db_url,
    similarity_threshold=0.2,
)

See it in action in this cookbook.