Using an Agent to iteratively search for information is called Agentic Search and the process of searching, reasoning and responding is known as Agentic RAG.

The model interprets your query, generates relevant keywords and searches its knowledge.

The Agent’s response is only as good as its search. Better search = Better responses

You can use semantic search, keyword search or hybrid search. We recommend using hybrid search with reranking for best in class agentic search.

Because the Agent is searching for the information it needs, this pattern is called Agentic Search and is becoming very popular with Agent builders.

Let’s build some examples to see Agentic Search in action.

​
Agentic RAG

When we add a knowledge base to an Agent, behind the scenes, we give the model a tool to search that knowledge base for the information it needs.

The Model generates a set of keywords and calls the search_knowledge_base() tool to retrieve the relevant information or few-shot examples.

Here’s a working example that uses Hybrid Search + Reranking:

You may remove the reranking step if you don’t need it.

agentic_rag.py
"""This cookbook shows how to implement Agentic RAG using Hybrid Search and Reranking.
1. Run: `pip install agno anthropic cohere lancedb tantivy sqlalchemy` to install the dependencies
2. Export your ANTHROPIC_API_KEY and CO_API_KEY
3. Run: `python cookbook/agent_concepts/agentic_search/agentic_rag.py` to run the agent
"""

from agno.agent import Agent
from agno.embedder.cohere import CohereEmbedder
from agno.knowledge.url import UrlKnowledge
from agno.models.anthropic import Claude
from agno.reranker.cohere import CohereReranker
from agno.vectordb.lancedb import LanceDb, SearchType

# Create a knowledge base, loaded with documents from a URL
knowledge_base = UrlKnowledge(
    urls=["https://docs.agno.com/introduction/agents.md"],
    # Use LanceDB as the vector database, store embeddings in the `agno_docs` table
    vector_db=LanceDb(
        uri="tmp/lancedb",
        table_name="agno_docs",
        search_type=SearchType.hybrid,
        embedder=CohereEmbedder(id="embed-v4.0"),
        reranker=CohereReranker(model="rerank-v3.5"),
    ),
)

agent = Agent(
    model=Claude(id="claude-3-7-sonnet-latest"),
    # Agentic RAG is enabled by default when `knowledge` is provided to the Agent.
    knowledge=knowledge_base,
    # search_knowledge=True gives the Agent the ability to search on demand
    # search_knowledge is True by default
    search_knowledge=True,
    instructions=[
        "Include sources in your response.",
        "Always search your knowledge before answering the question.",
        "Only include the output in your response. No other text.",
    ],
    markdown=True,
)

if __name__ == "__main__":
    # Load the knowledge base, comment after first run
    # knowledge_base.load(recreate=True)
    agent.print_response("What are Agents?", stream=True)

​
Agentic RAG with Reasoning

We can further improve the Agents search capabilities by giving it the ability to reason about the search results.

By adding reasoning, the Agent “thinks” first about what to search and then “analyzes” the results of the search.

Here’s an example of an Agentic RAG Agent that uses reasoning to improve the quality of the search results.

agentic_rag_reasoning.py
"""This cookbook shows how to implement Agentic RAG with Reasoning.
1. Run: `pip install agno anthropic cohere lancedb tantivy sqlalchemy` to install the dependencies
2. Export your ANTHROPIC_API_KEY and CO_API_KEY
3. Run: `python cookbook/agent_concepts/agentic_search/agentic_rag_with_reasoning.py` to run the agent
"""

from agno.agent import Agent
from agno.embedder.cohere import CohereEmbedder
from agno.knowledge.url import UrlKnowledge
from agno.models.anthropic import Claude
from agno.reranker.cohere import CohereReranker
from agno.tools.reasoning import ReasoningTools
from agno.vectordb.lancedb import LanceDb, SearchType

# Create a knowledge base, loaded with documents from a URL
knowledge_base = UrlKnowledge(
    urls=["https://docs.agno.com/introduction/agents.md"],
    # Use LanceDB as the vector database, store embeddings in the `agno_docs` table
    vector_db=LanceDb(
        uri="tmp/lancedb",
        table_name="agno_docs",
        search_type=SearchType.hybrid,
        embedder=CohereEmbedder(id="embed-v4.0"),
        reranker=CohereReranker(model="rerank-v3.5"),
    ),
)

agent = Agent(
    model=Claude(id="claude-3-7-sonnet-latest"),
    # Agentic RAG is enabled by default when `knowledge` is provided to the Agent.
    knowledge=knowledge_base,
    # search_knowledge=True gives the Agent the ability to search on demand
    # search_knowledge is True by default
    search_knowledge=True,
    tools=[ReasoningTools(add_instructions=True)],
    instructions=[
        "Include sources in your response.",
        "Always search your knowledge before answering the question.",
        "Only include the output in your response. No other text.",
    ],
    markdown=True,
)

if __name__ == "__main__":
    # Load the knowledge base, comment after first run
    # knowledge_base.load(recreate=True)
    agent.print_response(
        "What are Agents?",
        stream=True,
        show_full_reasoning=True,
        stream_intermediate_steps=True,
    )