PgVector Vector Database

The example uses OpenAI-backed embeddings or models. Set your key before running it:

export OPENAI_API_KEY="your-api-key"

Setup

uv pip install -U sqlalchemy psycopg pgvector pypdf openai agno

Run PgVector with Docker:

docker run -d \
  -e POSTGRES_DB=ai \
  -e POSTGRES_USER=ai \
  -e POSTGRES_PASSWORD=ai \
  -e PGDATA=/var/lib/postgresql \
  -v pgvolume:/var/lib/postgresql \
  -p 5532:5432 \
  --name pgvector \
  agnohq/pgvector:18

Example

Session storage uses the same PostgreSQL server through PostgresDb. The two requests reuse the Agent session, so the history tool can retrieve the earlier question.

agent_with_knowledge.py

from agno.agent import Agent
from agno.db.postgres import PostgresDb
from agno.models.openai import OpenAIResponses
from agno.knowledge.knowledge import Knowledge
from agno.vectordb.pgvector import PgVector, SearchType

db_url = "postgresql+psycopg://ai:ai@localhost:5532/ai"
knowledge_base = Knowledge(
    vector_db=PgVector(table_name="recipes", db_url=db_url, search_type=SearchType.hybrid),
)

if __name__ == "__main__":
    knowledge_base.insert(
        url="https://agno-public.s3.amazonaws.com/recipes/ThaiRecipes.pdf"
    )

    agent = Agent(
        model=OpenAIResponses(id="gpt-5.2"),
        knowledge=knowledge_base,
        db=PostgresDb(db_url=db_url),
        # Add a tool to read chat history.
        read_chat_history=True,
        markdown=True,
        # debug_mode=True,
    )
    agent.print_response("How do I make chicken and galangal in coconut milk soup", stream=True)
    agent.print_response("What was my last question?", stream=True)

Async Support ⚡

PgVector exposes async methods. Embedding and reader work can be awaited, while some ingestion writes still use synchronous SQLAlchemy and async search runs the synchronous search in a worker thread.

async_pgvector.py

import asyncio

from agno.agent import Agent
from agno.knowledge.knowledge import Knowledge
from agno.vectordb.pgvector import PgVector

db_url = "postgresql+psycopg://ai:ai@localhost:5532/ai"

vector_db = PgVector(table_name="recipes", db_url=db_url)

knowledge_base = Knowledge(
    vector_db=vector_db,
)

agent = Agent(knowledge=knowledge_base)

if __name__ == "__main__":
    # Load knowledge base asynchronously
    asyncio.run(knowledge_base.ainsert(
            url="https://agno-public.s3.amazonaws.com/recipes/ThaiRecipes.pdf"
        )
    )

    # Create and use the agent asynchronously
    asyncio.run(agent.aprint_response("How to make Tom Kha Gai", markdown=True))

Use ainsert() and aprint_response() to integrate with async application code. These methods do not make every database operation nonblocking; measure concurrency for your workload.

PgVector Params

Parameter	Type	Default	Description
`table_name`	`str`	-	Name of the table to store vector data.
`schema`	`str`	`"ai"`	Database schema name.
`name`	`Optional[str]`	`None`	Name of the vector database.
`description`	`Optional[str]`	`None`	Description of the vector database.
`id`	`Optional[str]`	`None`	ID of the vector database. Generated from `db_url` and `table_name` if not provided.
`db_url`	`Optional[str]`	`None`	Database connection URL.
`db_engine`	`Optional[Engine]`	`None`	SQLAlchemy database engine.
`embedder`	`Optional[Embedder]`	`None`	Embedder for creating embeddings. Defaults to `OpenAIEmbedder` if not provided.
`search_type`	`SearchType`	`vector`	Type of search to perform.
`vector_index`	`Union[Ivfflat, HNSW]`	`HNSW()`	Vector index configuration.
`distance`	`Distance`	`cosine`	Distance metric for vector comparisons.
`prefix_match`	`bool`	`False`	Enable prefix matching for full-text search.
`vector_score_weight`	`float`	`0.5`	Weight for vector similarity in hybrid search. Must be between 0 and 1.
`content_language`	`str`	`"english"`	Language for full-text search.
`schema_version`	`int`	`1`	Version of the database schema.
`reranker`	`Optional[Reranker]`	`None`	Reranker for reranking search results.
`create_schema`	`bool`	`True`	Create the database schema if it does not exist. Set to False if schema is managed externally.
`similarity_threshold`	`Optional[float]`	`None`	Minimum similarity score (0.0-1.0) to filter results.
`db`	`Optional[PostgresDb]`	`None`	Keyword-only: borrow a synchronous PostgreSQL database engine. Cannot be combined with `db_url` or `db_engine`.

Setup

Example

Async Support ⚡

PgVector Params

On this page