NoSQL databases

Key-value databases work when you require single-digit millisecond latency with well-defined access patterns (get/put by key). This type is good when building event-driven or microservices architectures, when using serverless with automatic scaling, and when global tables for multi-region applications are needed.

Use when migrating from MongoDB or need MongoDB compatibility, you need complex queries on document fields and nested data, your data has hierarchical nested structures, you require ACID transactions within documents, or you want a serverless option for variable workloads.

Use when your data is highly connected (social networks, fraud detection), you need to traverse relationships efficiently, you're building recommendation engines or knowledge graphs, you need both property graph (Gremlin) and RDF (SPARQL) support, or you require graph algorithms for analytics.

Use when you need Redis compatibility with durability guarantees, you're building microservices that need fast persistent data, you need both caching and primary database functionality, you require strong consistency with microsecond latency, or you want to simplify architecture by eliminating separate cache layers.

Use when you require sub-millisecond latency for frequently accessed data, you're implementing caching layers or session stores, you need Redis or Memcached compatibility, you want to reduce load on primary databases, or you're building real-time applications (gaming, chat).

Use when you need full-text search with relevance scoring, you're building search experiences or log analytics, you need real-time data visualization dashboards, you require complex aggregations and filtering, or you want a serverless option for cost optimization.