Youtube KnowledgeBase

Learn how to use YouTube video transcripts in your knowledge base.

The YouTubeKnowledgeBase iterates over a list of YouTube URLs, extracts the video transcripts, converts them into vector embeddings and loads them to a vector database.

Usage

We are using a local PgVector database for this example. Make sure it’s running

pip install bs4

knowledge_base.py

from agno.knowledge.youtube import YouTubeKnowledgeBase
from agno.vectordb.pgvector import PgVector

knowledge_base = YouTubeKnowledgeBase(
    urls=["https://www.youtube.com/watch?v=CDC3GOuJyZ0"],
    # Table name: ai.website_documents
    vector_db=PgVector(
        table_name="youtube_documents",
        db_url="postgresql+psycopg://ai:ai@localhost:5532/ai",
    ),
)

Then use the knowledge_base with an Agent:

agent.py

from agno.agent import Agent
from knowledge_base import knowledge_base

agent = Agent(
    knowledge=knowledge_base,
    search_knowledge=True,
)
agent.knowledge.load(recreate=False)

agent.print_response("Ask me about something from the knowledge base")

YouTubeKnowledgeBase also supports async loading.

pip install qdrant-client

We are using a local Qdrant database for this example. Make sure it’s running

async_knowledge_base.py

import asyncio

from agno.agent import Agent
from agno.knowledge.youtube import YouTubeKnowledgeBase, YouTubeReader
from agno.vectordb.qdrant import Qdrant

COLLECTION_NAME = "youtube-reader"

vector_db = Qdrant(collection=COLLECTION_NAME, url="http://localhost:6333")

knowledge_base = YouTubeKnowledgeBase(
    urls=[
        "https://www.youtube.com/watch?v=CDC3GOuJyZ0",
        "https://www.youtube.com/watch?v=JbF_8g1EXj4",
    ],
    vector_db=vector_db,
    reader=YouTubeReader(chunk=True),
)

agent = Agent(
    knowledge=knowledge_base,
    search_knowledge=True,
)

if __name__ == "__main__":
    # Comment out after first run
    asyncio.run(knowledge_base.aload(recreate=False))

    # Create and use the agent
    asyncio.run(
        agent.aprint_response(
            "What is the major focus of the knowledge provided in both the videos, explain briefly.",
            markdown=True,
        )
    )

Params

Parameter	Type	Default	Description
urls	List[str]	[]	URLs of the videos to read
reader	Optional[YouTubeReader]	None	A YouTubeReader that reads transcripts of the videos at the urls and converts them into Documents for the vector database.

YouTubeKnowledgeBase is a subclass of the AgentKnowledge class and has access to the same params.

Developer Resources

• View Sync loading Cookbook
• View Async loading Cookbook