Turbopuffer stands out for its excellent performance-to-cost ratio. Warm vector and hybrid (vector + BM25) queries are consistently fast, while the object-storage architecture keeps pricing dramatically lower than traditional vector DBs — strong ROI for both small and large workloads. The API and client libraries are clean and straightforward (good UI/UX for developers), integrations with common embedding and RAG stacks are simple, and onboarding is quick with clear docs. Native hybrid search and high-recall ANN make it very effective for AI/RAG use cases. Support has been responsive when needed. The only minor trade-off is higher latency on cold queries, which is expected given the design. Review collected by and hosted on G2.com.

The main things I dislike are the higher cold-query latency (expected with the object-storage design) and slightly higher write latency compared to pure in-memory vector databases. The ecosystem is also younger, so there are fewer third-party tools and integrations than with more established players. Index tuning is largely automatic, which is convenient but can feel limiting if you want fine-grained control. None of these are deal-breakers for me, and the cost + performance benefits more than make up for them. Review collected by and hosted on G2.com.