Case Study: Airtable achieves scalable, low-latency semantic search with Milvus from Zilliz

Airtable scales semantic search to 400k bases per cluster with Zilliz

Airtable, a leading digital operations platform, needed to build a scalable vector infrastructure to power semantic search and AI features within its product. Their challenge involved handling billions of embeddings for millions of independent customer bases while ensuring low-latency queries, high-throughput writes, and strict data privacy through a self-hosted solution. They evaluated several vendors and selected Zilliz's Milvus as their vector database.

The solution implemented with Zilliz's Milvus involved a partitioned architecture that provided strong isolation by mapping each customer base to its own physical partition. This allowed for high performance and simplified data deletion. Operationally, Airtable deployed Milvus on Kubernetes and utilized features like cold partition offloading to manage resources efficiently. The system successfully supports massive multi-tenancy, powers core AI features like Omni chat and linked record recommendations, and provides the low-latency, high-scale retrieval infrastructure required for Airtable's AI-powered platform.

Airtable