Search, Vector Search, and AI

How to design relevance, semantic retrieval, and AI application patterns with MongoDB services.

This guide covers the core concepts for full-text retrieval, semantic retrieval, and AI integration workflows.

Atlas Search

Atlas Search provides relevance-ranked full-text retrieval over operational data. Typical usage includes autocomplete, faceting, stemming, synonym expansion, and scoring customization.

Use Atlas Search when you need:

• Low-latency lexical search in user-facing experiences
• Rich query operators for relevance tuning
• Search indexes that evolve independently from primary schema changes

Vector Search Overview

Vector Search enables semantic retrieval by comparing embedding vectors using approximate nearest neighbor techniques. This is useful for retrieval-augmented generation (RAG), semantic recommendations, and multimodal search.

Typical workflow:

1. Generate embeddings for source documents.
2. Store vectors with metadata in MongoDB.
3. Query with a vector + metadata filters.
4. Rerank or post-filter before response generation.

AI Integrations

AI integrations connect MongoDB data to orchestration layers, model providers, and observability stacks. Production systems usually combine:

• Structured filtering in MongoDB
• Semantic candidate retrieval
• Model generation with citations and policy controls

Release Cadence

Atlas Search Release Notes

Track Search release changes when tuning analyzers, scoring, and index features to avoid regressions between environments.

Atlas Vector Search Release Notes

Track Vector Search release changes for index behavior, query operators, and scaling characteristics.

Related Guides

Atlas Build and Deploy

Atlas Security and Operations

Manual Core Capabilities

Development Overview