Chroma Hybrid Search

This example demonstrates how to use ChromaDB with hybrid search, which combines dense vector similarity search (semantic) with full-text search (keyword/lexical) using RRF fusion.

Hybrid search is useful when you want to:

Combine semantic understanding with exact keyword matching
Improve retrieval accuracy for queries with specific terms
Handle both conceptual and lexical search needs

The RRF algorithm fuses rankings from both search methods using:

1RRF(d) = sum(1 / (k + rank_i(d))) for each ranking i

Code

1import asyncio
2
3from kern.agent import Agent
4from kern.knowledge.knowledge import Knowledge
5from kern.vectordb.chroma import ChromaDb
6from kern.vectordb.search import SearchType
7
8# Create Knowledge Instance with ChromaDB using Hybrid Search
9knowledge = Knowledge(
10    name="Thai Recipes Knowledge Base",
11    description="Knowledge base for Thai recipes with hybrid search (RRF fusion)",
12    vector_db=ChromaDb(
13        collection="thai_recipes_hybrid",
14        path="tmp/chromadb_hybrid",
15        persistent_client=True,
16        # Enable hybrid search - combines vector similarity with keyword matching using RRF
17        search_type=SearchType.hybrid,
18        # RRF (Reciprocal Rank Fusion) constant - controls ranking smoothness.
19        # Higher values (e.g., 60) give more weight to lower-ranked results,
20        # Lower values make top results more dominant. Default is 60 (per original RRF paper).
21        hybrid_rrf_k=60,
22    ),
23)
24
25# Load content into the knowledge base
26asyncio.run(
27    knowledge.ainsert(
28        name="Thai Recipes",
29        url="https://kern-public.s3.amazonaws.com/recipes/ThaiRecipes.pdf",
30        metadata={"doc_type": "recipe_book", "cuisine": "thai"},
31    )
32)
33
34# Create an agent with the hybrid search knowledge base
35agent = Agent(
36    knowledge=knowledge,
37    search_knowledge=True,
38    instructions="You are a helpful Thai cooking assistant. Use the knowledge base to answer questions about Thai recipes.",
39)
40
41# Hybrid search will:
42# 1. Find semantically similar documents (via dense embeddings)
43# 2. Find documents containing query keywords (via FTS)
44# 3. Fuse results using RRF for optimal ranking
45agent.print_response("What are the ingredients for Massaman curry?", markdown=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 chromadb openai kern-ai

Run Agent

1python cookbook/08_knowledge/vector_db/chroma_db/chroma_db_hybrid_search.py