SurrealDB vs. MongoDB

Architecture

Document-first with secondary views, full-text search, and vectors requiring duplicated data and separate async pipelines.

Unified, distributed multi-model engine. Query execution, indexing, storage, and transactions operate inside a single system.

Models

Document-first. Graph traversal support is limited. Relational semantics are not planner-driven.

Native support for document, relational, graph, key-value, time-series, vector, full-text search, and geospatial.

Transactional consistency

ACID transactions are limited, costly under contention, and don't extend to search, vectors, or graph traversal.

ACID transactions across documents, relational joins, graph traversal, vector search, and full-text search.

Query execution

Queries split into multiple stages, with $search required first. No unified planner across document, search, vector, and graph operations.

Single declarative query language with a unified execution plan. All retrieval primitives are co-planned.

Indexing

Compound indexes may include only one array field. Deeply nested or multi-array document models cannot be efficiently indexed.

Indexing across arrays, nested fields, vectors, and relationships without compound multikey restrictions.

Temporal capabilities

No temporal graph querying. Time filters cannot participate directly in traversal semantics.

Temporal querying is first-class. Time constraints participate directly in graph traversal.

Pricing

Costs increase with sharding, coordination overhead, Atlas dependency, and duplicated subsystems.

Unified engine eliminates the need for separate search, vector, and graph systems. Costs scale linearly.