# MongoDB Atlas Search and Vector Search > ```python Parent: [MongoDB Atlas](https://llms-explorer.com/tree/mongodb-atlas/) · 12 facets · 34 facts · page: https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/ ## Pagination with Atlas Search - For deterministic pagination (not skip/limit): — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#pagination-with-atlas-search) ## Atlas Vector Search - See mongodb-atlas-vector-search for the complete reference. Key summary: — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#atlas-vector-search) ## Hybrid Search ($rankFusion) - Combine Atlas Search (BM25) with Vector Search (HNSW) results: — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#hybrid-search-rankfusion) ## Auto Embedding (Voyage AI) - Skip the external embedding pipeline: — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#auto-embedding-voyage-ai) ## Search Node Architecture - For production, use dedicated Search Nodes to isolate Atlas Search / Vector Search workloads from OLTP: — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#search-node-architecture) - S20_HIGHCPU_NVME, S30_HIGHCPU_NVME, etc. — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#search-node-architecture) - Zero-downtime migration: enable Search Nodes → Atlas replications in background → automatic traffic switch — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#search-node-architecture) - See mongodb-atlas-search-nodes for sizing guide — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#search-node-architecture) ## When to Use Atlas Search vs Text Indexes - Always prefer Atlas Search for production full-text search. Text indexes are only appropriate for very small collections or simple dev prototypes. — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#when-to-use-atlas-search-vs-text-indexes) ## Anti-Patterns - Dynamic true + all fields queried: Dynamic mapping indexes everything, making the index large. Use explicit mappings for production. — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#anti-patterns) - No search node for production Vector Search: Resource contention with OLTP causes latency spikes — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#anti-patterns) - numCandidates too low: numCandidates: limit = minimal recall; use 10-20× limit — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#anti-patterns) - Missing filter fields in vector index: filter in $vectorSearch requires the field to be declared as type: "filter" in the index definition; otherwise falls back to post-filter (much less efficient) — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#anti-patterns) ## References - Atlas Search Documentation — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#references) - $search Aggregation Stage — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#references) - $rankFusion — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#references) - Atlas Vector Search — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#references) - Voyage AI Auto-Embedding — [source](https://llms-explorer.com/sources/mdb-context-hub/mongodb-search-ai/#references) ## Where this helps - Building a hybrid search feature that needs to merge keyword relevance (BM25) with semantic similarity (vector search) into one ranked result set. — [source](https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/) *(AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.)* - Deciding whether a production Vector Search workload needs dedicated Search Nodes to avoid resource contention with the OLTP cluster. — [source](https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/) *(AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.)* - Adopting automated embedding, via Voyage AI models through autoEmbed, to skip maintaining an external embedding pipeline. — [source](https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/) *(AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.)* - Implementing deterministic pagination for a search UI where skip/limit pagination would be too slow or unstable at depth. — [source](https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/) *(AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.)* ## Project ideas - Build a $rankFusion query that merges Atlas Search (BM25) and Atlas Vector Search (HNSW) results into a single hybrid-ranked response. — [source](https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/) *(AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.)* - Migrate an external embedding pipeline to Atlas's built-in autoEmbed field type backed by Voyage AI, and compare operational complexity before and after. — [source](https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/) *(AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.)* - Provision dedicated Search Nodes for a production Vector Search workload and measure query latency before and after isolating it from OLTP. — [source](https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/) *(AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.)* - Build a vector index with an explicitly declared filter-type field so filters in $vectorSearch use the index instead of falling back to a much slower post-filter. — [source](https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/) *(AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.)* ## Common mistakes - Leaving dynamic mapping enabled with every field queried, producing an unnecessarily large index instead of using explicit mappings in production. — [source](https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/) *(AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.)* - Running production Vector Search without a dedicated Search Node, letting it contend for resources with OLTP and causing latency spikes. — [source](https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/) *(AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.)* - Setting numCandidates too close to limit, which minimizes recall — a 10–20× multiple of limit is the recommended starting point. — [source](https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/) *(AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.)* - Omitting a field from the vector index's filter-type declaration and then filtering on it in $vectorSearch anyway, which silently falls back to a much less efficient post-filter. — [source](https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/) *(AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.)* ## Known issues - This pack's own summary field is malformed, a stray code-fence marker instead of a real summary, a sign the source content it was generated from may have been imperfectly extracted — verify details against the fuller mongodb-atlas-search and mongodb-atlas-vector-search references before relying on this pack alone. — [source](https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/) *(AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.)* - Auto-embedding via Voyage AI depends on a specific external model family (voyage-4-large, voyage-4, voyage-4-lite, voyage-code-3); switching providers or models later requires re-embedding existing data. — [source](https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/) *(AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.)* - Search Node migration to production is described as zero-downtime via background replication and automatic traffic switch, but that migration still needs to be planned and executed deliberately, not assumed automatic on enabling the feature. — [source](https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/) *(AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.)* - Text indexes are explicitly called out as appropriate only for very small collections or simple dev prototypes — they are not a long-term substitute for Atlas Search. — [source](https://llms-explorer.com/tree/mongodb-atlas-search-and-vector-search/) *(AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.)* ## Context files - [MongoDB Atlas Search and Vector Search](https://llms-explorer.com/downloads/sources/mdb-context-hub/mongodb-search-ai.md)