Agentic RAG with Reranking

Combine agentic search, hybrid retrieval, and reranking for high-quality responses.

This example combines three techniques for optimal retrieval:

Agentic RAG: Agent decides when to search the knowledge base
Hybrid search: Combines vector similarity with keyword matching
Reranking: Reorders results using a dedicated ranking model

1from kern.agent import Agent
2from kern.knowledge.knowledge import Knowledge
3from kern.knowledge.embedder.cohere import CohereEmbedder
4from kern.knowledge.reranker.cohere import CohereReranker
5from kern.models.anthropic import Claude
6from kern.vectordb.lancedb import LanceDb, SearchType
7
8knowledge = Knowledge(
9    vector_db=LanceDb(
10        uri="tmp/lancedb",
11        table_name="docs",
12        search_type=SearchType.hybrid,
13        embedder=CohereEmbedder(id="embed-v4.0"),
14        reranker=CohereReranker(model="rerank-v3.5"),
15    ),
16)
17
18agent = Agent(
19    model=Claude(id="claude-sonnet-4-5"),
20    knowledge=knowledge,
21    search_knowledge=True,
22)

Why Combine These Techniques

Technique	What It Does
Agentic RAG	Agent searches only when needed, can reformulate queries
Hybrid search	Catches both semantic matches and exact terms
Reranking	Uses a dedicated model to reorder results by relevance

Together, these provide better retrieval accuracy than any single technique alone.

How Reranking Works

After hybrid search returns initial results, the reranker:

Takes the query and candidate documents
Scores each document for relevance using a cross-encoder model
Reorders results so the most relevant appear first

Cohere's rerank-v3.5 is trained specifically for this task and significantly improves result quality.

Example

1import asyncio
2
3from kern.agent import Agent
4from kern.knowledge.embedder.cohere import CohereEmbedder
5from kern.knowledge.knowledge import Knowledge
6from kern.knowledge.reranker.cohere import CohereReranker
7from kern.models.anthropic import Claude
8from kern.vectordb.lancedb import LanceDb, SearchType
9
10# Create knowledge base with hybrid search and reranking
11knowledge = Knowledge(
12    vector_db=LanceDb(
13        uri="tmp/lancedb",
14        table_name="agno_docs",
15        search_type=SearchType.hybrid,
16        embedder=CohereEmbedder(id="embed-v4.0"),
17        reranker=CohereReranker(model="rerank-v3.5"),
18    ),
19)
20
21# Load content
22asyncio.run(
23    knowledge.ainsert(url="https://kern.ndx.rocks/introduction/agents.md")
24)
25
26# Create agent with knowledge
27agent = Agent(
28    model=Claude(id="claude-sonnet-4-20250514"),
29    knowledge=knowledge,
30    search_knowledge=True,
31    instructions=[
32        "Search your knowledge before answering.",
33        "Include sources in your response.",
34    ],
35    markdown=True,
36)
37
38agent.print_response("What are Agents?", stream=True)

Usage

Set up your virtual environment

1uv venv --python 3.12
2source .venv/bin/activate

1uv venv --python 3.12
2.venv\Scripts\activate

Install dependencies

1uv pip install -U kern-ai anthropic cohere lancedb tantivy sqlalchemy

Export your API keys

1export ANTHROPIC_API_KEY=your_anthropic_api_key_here
2export CO_API_KEY=your_cohere_api_key_here

Run Agent

1python agentic_rag.py

Configuration Options

Different Rerankers

1# Cohere
2from kern.knowledge.reranker.cohere import CohereReranker
3reranker = CohereReranker(model="rerank-v3.5")
4
5# Add to vector database
6vector_db = LanceDb(
7    uri="tmp/lancedb",
8    table_name="docs",
9    search_type=SearchType.hybrid,
10    reranker=reranker,
11)

Adjusting Results

1knowledge = Knowledge(
2    vector_db=vector_db,
3    max_results=10,  # Number of results to return after reranking
4)

Next Steps

magnifying-glass

Hybrid Search

Learn more about combining search types

code

Embedders

Choose the right embedding model