Book a Demo Start Instantly

In the realm of artificial intelligence, efficient similarity search and clustering are crucial for handling large datasets. FAISS (Facebook AI Similarity Search) and TiDB Vector Search together offer a powerful combination for implementing semantic search and vector similarity tasks. This article delves into how FAISS can be integrated with TiDB Vector Search to enhance AI applications.

What is FAISS?

FAISS is a library developed by Facebook AI Research that allows for efficient similarity search and clustering of dense vectors. It is designed to handle large-scale datasets, making it an excellent choice for applications requiring fast and accurate nearest neighbor search.

What is TiDB Vector Search?

TiDB Vector Search allows for storing and searching vector embeddings within TiDB, a distributed SQL database. By combining FAISS with TiDB, you can leverage the strengths of both: FAISS for high-performance vector indexing and search, and TiDB for robust data storage and management.

Integrating FAISS with TiDB Vector Search

Step 1: Setting Up Your Environment

To begin, ensure you have the following prerequisites:

  • Python >= 3.6
  • TiDB Serverless Cluster with vector support (Sign up at
  • FAISS library

Step 2: Install Required Libraries

Install FAISS and other necessary libraries:

pip install faiss-cpu mysql-connector-python

Step 3: Connecting to TiDB and Creating Tables

First, connect to your TiDB cluster and create the necessary tables:

import mysql.connector
import faiss
import numpy as np

# Connect to TiDB
conn = mysql.connector.connect(
cursor = conn.cursor()

# Create table to store embeddings
    doc TEXT,
    embedding BLOB

Step 4: Generating and Storing Embeddings

Use a model like Llama 3 to generate embeddings and store them in TiDB:

from openai import OpenAI

client = OpenAI(api_key='your_openai_api_key')
documents = ["Sample text 1", "Sample text 2", "Sample text 3"]
embeddings = [r.embedding for r in client.embeddings.create(input=documents, model="text-embedding-3-small").data]

# Store embeddings in TiDB
for doc, emb in zip(documents, embeddings):
    cursor.execute("INSERT INTO embeddings (doc, embedding) VALUES (%s, %s)", (doc, emb.tobytes()))

Step 5: Building FAISS Index

Retrieve embeddings from TiDB and build a FAISS index:

# Fetch embeddings from TiDB
cursor.execute("SELECT id, embedding FROM embeddings")
rows = cursor.fetchall()

# Convert embeddings to numpy array
embedding_matrix = np.array([np.frombuffer(row[1], dtype=np.float32) for row in rows])
ids = np.array([row[0] for row in rows])

# Create FAISS index
dimension = embedding_matrix.shape[1]
index = faiss.IndexFlatL2(dimension)
index.add_with_ids(embedding_matrix, ids)

Step 6: Performing Similarity Search with FAISS

Use FAISS to perform similarity searches:

# Example query vector
query_vector = client.embeddings.create(input="Query text", model="text-embedding-3-small").data[0].embedding

# Convert query vector to numpy array
query_vector = np.array(query_vector, dtype=np.float32).reshape(1, -1)

# Search for nearest neighbors
D, I =, k=5)  # k is the number of nearest neighbors to retrieve

# Fetch and display results
for i in I[0]:
    cursor.execute("SELECT doc FROM embeddings WHERE id = %s", (i,))
    result = cursor.fetchone()
    if result:


By integrating FAISS with TiDB Vector Search, you can create a powerful system capable of handling large-scale similarity search tasks with high performance and accuracy. This combination leverages FAISS’s efficient vector indexing and TiDB’s robust data management, making it ideal for advanced AI applications. Get started with TiDB Serverless today by visiting TiDB Cloud and explore its potential for your AI projects.

Last updated June 26, 2024

Spin up a Serverless database with 25GiB free resources.

Start Now