"""Demonstrates knowledge isolation with isolate_vector_search flag.
When multiple Knowledge instances share the same vector database, you can use
the `isolate_vector_search` flag to ensure each instance only searches its own data.
Behavior:
- isolate_vector_search=False (default): Searches ALL vectors in the database.
This is backwards-compatible with existing data that doesn't have linked_to metadata.
- isolate_vector_search=True: Only searches vectors that have matching linked_to metadata.
Documents inserted with this flag will have linked_to set to the Knowledge instance name.
Searches will filter to only return documents with matching linked_to.
IMPORTANT: If you have existing production data and want to enable isolation, you will
need to re-index your data with isolate_vector_search=True to add the linked_to metadata.
Existing documents without linked_to metadata will NOT be found when isolation is enabled.
Run: `python cookbook/07_knowledge/basic_operations/async/05_isolate_vector_search.py`
"""
import asyncio
from agno.agent import Agent
from agno.db.postgres.postgres import PostgresDb
from agno.knowledge.knowledge import Knowledge
from agno.vectordb.pgvector import PgVector
# Shared database connections
contents_db = PostgresDb(
db_url="postgresql+psycopg://ai:ai@localhost:5532/ai",
knowledge_table="knowledge_contents",
)
vector_db = PgVector(
table_name="vectors",
db_url="postgresql+psycopg://ai:ai@localhost:5532/ai",
)
# -----------------------------------------------------------------------------
# Example 1: Default behavior (isolate_vector_search=False)
# Searches across ALL vectors in the database, regardless of which Knowledge
# instance inserted them. This is the default for backwards compatibility.
# -----------------------------------------------------------------------------
knowledge_shared = Knowledge(
name="Shared Knowledge",
description="This knowledge instance searches all vectors",
vector_db=vector_db,
contents_db=contents_db,
# isolate_vector_search=False is the default
)
# -----------------------------------------------------------------------------
# Example 2: Isolated behavior (isolate_vector_search=True)
# Only searches vectors that were inserted by this Knowledge instance.
# Documents are tagged with linked_to metadata during insert.
# -----------------------------------------------------------------------------
knowledge_isolated = Knowledge(
name="Isolated Knowledge",
description="This knowledge instance only searches its own vectors",
vector_db=vector_db,
contents_db=contents_db,
isolate_vector_search=True, # Enable isolation
)
async def main():
# Insert a document with isolation enabled
# This document will have linked_to="Isolated Knowledge" in its metadata
await knowledge_isolated.ainsert(
name="CV",
path="cookbook/07_knowledge/testing_resources/cv_1.pdf",
metadata={"user_tag": "Engineering Candidates"},
)
# Agent using isolated knowledge - only finds documents from this instance
agent_isolated = Agent(
name="Isolated Agent",
knowledge=knowledge_isolated,
search_knowledge=True,
debug_mode=True,
)
print("--- Agent with isolate_vector_search=True ---")
print("Only searches vectors with linked_to='Isolated Knowledge'")
agent_isolated.print_response(
"What skills does Jordan Mitchell have?",
markdown=True,
)
# Agent using shared knowledge - finds ALL documents in the vector db
agent_shared = Agent(
name="Shared Agent",
knowledge=knowledge_shared,
search_knowledge=True,
debug_mode=True,
)
print("--- Agent with isolate_vector_search=False (default) ---")
print("Searches all vectors in the database")
agent_shared.print_response(
"What skills does Jordan Mitchell have?",
markdown=True,
)
if __name__ == "__main__":
asyncio.run(main())
