Skip to main content
Knowledge is the retrieval primitive: a vector index with an optional keyword index and optional reranker. Pal and Dash both use it heavily. 
from agno.knowledge import Knowledge
from agno.vector_db.pgvector import PgVector

dash_knowledge = Knowledge(
    vector_db=PgVector(
        table_name="dash_knowledge",
        db_url=DB_URL,
        search_type="hybrid",       # vector + BM25
    ),
)

dash = Agent(
    knowledge=dash_knowledge,
    add_knowledge_to_context=True,  # auto-search before each run
    search_knowledge=True,          # also expose as a tool
)

​
Loading content

Three ways to put content in: 
# From a directory
dash_knowledge.add_content_from_path("knowledge/tables/")

# From a URL
pal_knowledge.add_content_from_url("https://example.com/article")

# Programmatically
dash_knowledge.add_content(
    name="MRR definition",
    content="MRR is sum of active subscriptions excluding trials.",
    metadata={"category": "business_rules"},
)
Demo OS loads via scripts: 
python -m agents.dash.scripts.load_knowledge
python -m agents.pal.scripts.load_knowledge
Re-run with --recreate to rebuild from scratch. Without it, content is upserted by primary key.

​
Chunking and embedding

By default, Knowledge chunks long content into ~500-token segments and embeds each chunk with text-embedding-3-small. Override with: 
from agno.embedder.openai import OpenAIEmbedder
from agno.knowledge.chunking.text import TextChunkingStrategy

dash_knowledge = Knowledge(
    vector_db=PgVector(...),
    embedder=OpenAIEmbedder(id="text-embedding-3-large"),
    chunking_strategy=TextChunkingStrategy(chunk_size=1000, overlap=100),
)
Other chunking strategies live under agno.knowledge.chunking.*: by markdown headers, by code structure, by recursive token count, by semantic boundaries. search_type="hybrid" runs both:
IndexCatches
Vector (semantic)“different words for the same idea”
BM25 (keyword)“find the doc that mentions this exact term”
Results from both get merged with reciprocal rank fusion. Hybrid almost always beats either alone.

​
Metadata filtering

Metadata attached at ingest time becomes a filter at query time: 
# Ingest
dash_knowledge.add_content(
    name="MRR definition",
    content="...",
    metadata={"category": "business_rules", "team": "finance"},
)

# Retrieve only finance team rules
dash_knowledge.search(
    query="how do we calculate MRR?",
    filters={"team": "finance"},
)
Useful for multi-tenant agents (filter by tenant_id) or topic scoping (filter by category).

​
When the model gets the chunks

With add_knowledge_to_context=True:
  1. User message arrives.
  2. AgentOS runs knowledge.search(message) automatically.
  3. Top-k chunks get inserted into the system prompt.
  4. The model answers with the chunks visible.
With search_knowledge=True: The agent gets a search_knowledge_base(query) tool. The model decides when to call it. Useful for follow-up retrieval mid-run. Both flags are common to set together. Auto-search hits first, the tool catches “I need to look up something else” cases.

​
Reranking

For larger knowledge bases, add a reranker: 
from agno.rerank.cohere import CohereReranker

dash_knowledge = Knowledge(
    vector_db=PgVector(...),
    reranker=CohereReranker(model="rerank-3.5"),
)
The vector DB returns the top-50, the reranker scores them, and the top-10 reach the model. This two-stage retrieval (cast wide, rerank tight) is the standard production setup.

​
See it in action

@Pal what do I know about transformers?
@Dash what's the right way to count active subscriptions?
@Pal which articles have I ingested about RAG?
@Dash show me a query for MRR by plan
Source: agents/dash/, Knowledge docs

​
Next

Memory →