ScyllaDB Vector Search is now Generally Available (GA)

Release Notes
, , ,
1

We are thrilled to announce that ScyllaDB Vector Search is now Generally Available (GA) and production-ready in ScyllaDB Cloud.

Many organizations are embracing real-time AI, but they often encounter significant challenges with existing vector databases - whether standalone or as an extension to a NoSQL offering. These solutions frequently prove too complex and expensive to manage at scale, while also failing to deliver the low latency and high throughput required for critical business needs and a superior customer experience (UX).

ScyllaDB Vector Search directly addresses these issues by integrating millisecond-latency vector retrieval into the core database. This makes it an ideal, high-throughput solution for applications such as Retrieval-Augmented Generation (RAG), fraud detection, and real-time recommendation engines.

Key Highlights

  • Massive Scale & Speed: Leverages our close-to-bare-metal, shard-per-core architecture together with a uniquely integrated, uSearch library which is Rust-based, ultra-fast VSS Index nodes so that we are able to deliver our known high-scale, high throughput, predictable sub-millisecond P99 latencies also for VSS queries/inferences at scale and with high recalls.

  • Unified Architecture: You can now store vector embeddings and structured attributes in the same database, getting both regular and VSS queries at the speed of light.

  • Independent Scaling: The design allows storage and indexing responsibilities to scale independently, building optimized, low-cost solutions. The VSS nodes ensure there is no “noisy neighbor” effects when VSS workload is done in parallel to other workloads.

  • Low TCO: By consolidating your stack, and utilizing the fastest QPS per vCPU VSS solution, you can achieve significant TCO savings compared to maintaining separate vector solutions.

Get Started

We are eager to hear your feedback as you begin using this feature for your AI and ML workloads.

1 Like