agno-knowledge

name: agno-knowledge description: | Build knowledge bases for Agno agents using vector databases, embedders, and document readers. Covers ChromaDB, PgVector, LanceDB, document loading, chunking strategies, and hybrid search. Trigger this skill when: importing agno.knowledge, creating Knowledge instances, adding documents to agents, configuring vector databases, or asking "how do I add a knowledge base?" license: Apache-2.0 metadata: version: "1.0.0" author: agno-team tags: ["knowledge", "rag", "vector-db", "embeddings", "agno"]

Build Agno Knowledge Bases

Use agno.knowledge.knowledge.Knowledge with a vector database to give agents searchable document collections. Install with pip install agno.

Quick Start

from agno.agent import Agent
from agno.knowledge.knowledge import Knowledge
from agno.knowledge.embedder.google import GeminiEmbedder
from agno.vectordb.chroma import ChromaDb
from agno.vectordb.search import SearchType

knowledge = Knowledge(
    vector_db=ChromaDb(
        collection="docs",
        path="tmp/chromadb",
        persistent_client=True,
        search_type=SearchType.hybrid,
        embedder=GeminiEmbedder(id="gemini-embedding-001"),
    ),
)

# Load content
knowledge.insert(url="https://docs.agno.com/introduction.md", skip_if_exists=True)

agent = Agent(
    model="openai:gpt-4o",
    knowledge=knowledge,
    search_knowledge=True,
    markdown=True,
)

agent.print_response("What is Agno?", stream=True)

Core Knowledge Parameters

Knowledge(
    vector_db=ChromaDb(...),      # Required: vector database backend
    max_results=5,                # Max documents returned per search
    contents_db=SqliteDb(...),    # Optional: track loaded content
)

Vector Database Backends

ChromaDB (Local, no server needed)

from agno.vectordb.chroma import ChromaDb
from agno.knowledge.embedder.google import GeminiEmbedder
from agno.vectordb.search import SearchType

vector_db = ChromaDb(
    collection="my_docs",
    path="tmp/chromadb",
    persistent_client=True,
    search_type=SearchType.hybrid,
    embedder=GeminiEmbedder(id="gemini-embedding-001"),
)

PgVector (PostgreSQL)

from agno.vectordb.pgvector import PgVector
from agno.knowledge.embedder.openai import OpenAIEmbedder

vector_db = PgVector(
    table_name="documents",
    db_url="postgresql+psycopg://user:pass@localhost:5432/mydb",
    search_type=SearchType.hybrid,
    embedder=OpenAIEmbedder(id="text-embedding-3-small"),
)

LanceDB (Local, columnar)

from agno.vectordb.lancedb import LanceDb
from agno.knowledge.embedder.google import GeminiEmbedder

vector_db = LanceDb(
    uri="tmp/lancedb",
    table_name="documents",
    embedder=GeminiEmbedder(id="gemini-embedding-001"),
)

Embedders

# Google
from agno.knowledge.embedder.google import GeminiEmbedder
embedder = GeminiEmbedder(id="gemini-embedding-001")

# OpenAI
from agno.knowledge.embedder.openai import OpenAIEmbedder
embedder = OpenAIEmbedder(id="text-embedding-3-small")

# Ollama (local)
from agno.knowledge.embedder.ollama import OllamaEmbedder
embedder = OllamaEmbedder(id="nomic-embed-text")

Loading Documents

From URLs

knowledge.insert(url="https://example.com/doc.md", skip_if_exists=True)

From Local Files (PDF, DOCX, Text)

from agno.knowledge.reader.pdf import PDFReader
from agno.knowledge.reader.docx import DocxReader
from agno.knowledge.reader.text import TextReader

# PDF
knowledge.insert(
    path="docs/report.pdf",
    reader=PDFReader(),
    skip_if_exists=True,
)

# DOCX
knowledge.insert(
    path="docs/spec.docx",
    reader=DocxReader(),
    skip_if_exists=True,
)

# Plain text / markdown
knowledge.insert(
    path="docs/readme.md",
    reader=TextReader(),
    skip_if_exists=True,
)

From Directories

knowledge.insert(
    path="docs/",
    reader=PDFReader(),
    skip_if_exists=True,
)

Search Types

from agno.vectordb.search import SearchType

# Semantic similarity (default)
SearchType.vector

# Keyword-based
SearchType.keyword

# Combined semantic + keyword
SearchType.hybrid

Chunking

Control how documents are split into searchable chunks:

from agno.knowledge.chunking.fixed import FixedChunking
from agno.knowledge.chunking.semantic import SemanticChunking

# Fixed-size chunks
knowledge = Knowledge(
    vector_db=vector_db,
    chunking=FixedChunking(chunk_size=500, overlap=50),
)

# Semantic chunking (splits at natural boundaries)
knowledge = Knowledge(
    vector_db=vector_db,
    chunking=SemanticChunking(),
)

Knowledge with Content Tracking

Track which documents have been loaded to avoid re-processing:

from agno.db.sqlite import SqliteDb

db = SqliteDb(db_file="tmp/agents.db")

knowledge = Knowledge(
    vector_db=vector_db,
    contents_db=db,
    max_results=5,
)

Agent with Knowledge

agent = Agent(
    model="openai:gpt-4o",
    knowledge=knowledge,
    search_knowledge=True,       # Agent searches automatically
    # OR
    # search_knowledge=False,    # Agent uses knowledge tools manually
    markdown=True,
)

Knowledge Tools

Give agents explicit control over knowledge search:

from agno.tools.knowledge import KnowledgeTools

agent = Agent(
    model="openai:gpt-4o",
    knowledge=knowledge,
    tools=[KnowledgeTools()],
    search_knowledge=False,  # Let the agent decide when to search
)

Anti-Patterns

Don't skip skip_if_exists=True on insert — you'll re-index documents on every run
Don't forget an embedder — the vector DB needs one to generate embeddings
Don't use SearchType.keyword alone — hybrid gives better results in most cases
Don't load huge documents without chunking — they won't fit in context windows
Don't hardcode embedder models — match the embedder to your deployment constraints