AI Search for Agents: Announcing Automated Embedding in MongoDB Atlas

Automated Embedding, now in Public Preview on MongoDB Atlas, offers an advanced solution for embedding and vector search, eliminating the need for parallel embedding pipelines and manual backfill jobs. Utilizing Voyage AI embedding models, it ensures near real-time synchronization by re-embedding only when indexed fields change, thus maintaining up-to-date and reliable data for agents. The system enhances performance by separating inference into two mechanisms optimized for throughput and latency, preventing one workload from impacting another. Additionally, configurable quantization and dynamic batching allow efficient storage and indexing, supporting scalability to hundreds of millions of documents without bottlenecks. Feature engineering is simplified to a query-language exercise with automated embeddings on views, enabling high retrieval accuracy without maintaining separate search index documents. This integrated approach transforms search from an integration project into a fundamental component of agent applications, offering fast, compliant, and affordable retrieval solutions.