No cluster to size, no idle capacity to pay for -- storage and query-based pricing for embeddings, from fully managed platforms to object-storage-native engines and Postgres-integrated options. Curated and link-checked, not just listed.
Fully managed vector database built for AI, autoscaling by design
Pinecone's serverless architecture separates storage from compute and scales indexes automatically -- writes are instantly searchable and performance stays consistent regardless of data size. Pricing follows storage and query volume rather than provisioned capacity, so idle indexes cost close to nothing.
Visit site →Disk-based serverless vector search, pay only for what you use
Built for high-performance search at hundreds of millions of vectors, Upstash Vector uses a disk-based algorithm for cost-effective storage plus a metadata store with SQL support for flexible filtering. True consumption pricing -- from the same team behind Upstash's serverless Redis and Kafka.
Visit site →Vector + full-text search built directly on object storage
Rather than running a persistent cluster, turbopuffer fronts S3-compatible object storage with a memory/SSD cache layer -- eliminating dedicated infrastructure while still handling 1T+ document workloads with sub-10ms query latency. Positioned as roughly 10x cheaper than cluster-based alternatives.
Visit site →Serverless multimodal lakehouse built on the open-source Lance format
LanceDB unifies vector search, full-text search, and hybrid retrieval in a single table, built on the open-source columnar Lance format for petabyte-scale data curation. Runs serverless against object storage, so there's no separate index infrastructure to size or manage alongside your data lake.
Visit site →AI-native vector database, open-source core with a usage-based cloud tier
Weaviate stores, indexes, and searches high-dimensional vectors at scale, with both a self-hosted open-source path and Weaviate Cloud offering a free tier plus usage-based pricing. A good fit when you want the option to self-host later without a migration.
Visit site →Open-source Rust vector engine; serverless tier explicitly in progress
Qdrant is an open-source vector search engine written in Rust, available self-hosted, at the edge, or via Qdrant Cloud's fully managed offering with auto-sharding. Worth flagging plainly: as of this review, Qdrant's own site lists a dedicated serverless tier as "coming soon" rather than shipped today.
Visit site →Serverless vector search on Apache Cassandra, now part of IBM
Astra DB delivers vector search on top of Apache Cassandra with a pay-as-you-go serverless pricing model and near-zero latency at scale. DataStax was acquired into IBM's portfolio; the product page also covers Hyper-Converged Database (HCD) for teams that need an on-premises deployment path alongside the cloud service.
Visit site →Vector embeddings inside your existing serverless Postgres database
Supabase's Vector module adds embeddings storage, indexing, and search directly inside Postgres via pgvector, with out-of-the-box support for OpenAI, Hugging Face, and other embedding sources. The advantage is architectural, not just pricing: vectors live alongside your relational data instead of a separate store.
Visit site →