Open-source vector database startup Qdrant Solutions GmbH today announced three new enterprise-grade capabilities on its cloud service to address performance, availability and compliance requirements for artificial intelligence and always-on retrieval needs. 

The new services are graphics processing unit-accelerated indexing, Multi-AZ clusters and audit logging. 

As more AI systems become mission-critical for the agentic era and enterprise companies begin to adopt retrieval-augmented generation, also known as RAG, vector search has become the go-to for data retrieval. Vector databases form the engine of semantic understanding for chatbots and AI agents, providing real-time information, reducing hallucinations and increasing accuracy. 

“GPUs aren’t just for model inference. They’re for indexing too,” said co-founder and Chief Executive Andre Zayarni. 

Indexes create specialized scaffolds to organize data within vector databases to enable lightning-fast similarity searches across vast oceans of information. By clustering similar vectors using algorithms such as Hierarchical Navigable Small World and Inverted File, they replace the slow, brute-force comparisons used in classical databases. 

They are foundational to high-speed lookups used to deliver data at the blistering speeds needed to make AI feel like human response speeds. Without them, an AI chatbot or agent would be slow and inarticulate. They are used across numerous systems such as recommendation and search engines. 

In addition to speeding up indexing with GPUs, Qdrant also added resilience with Multi-AZ clusters, which maintain copies across three availability zones within a single region using replication. No failover. If a single instance goes offline, reads and writes continue from the surviving zones without any delay and no customer action is required. 

Audit logging captures all activity across the Qdrant application programming interface, including queries, deletes, collection management and snapshot management. It provides structured JSON with API user key attribution, timestamps and other metadata, providing a full audit trail. Retention is configurable.  

For long-term storage, customers can download logs for archival and compliance purposes. 

Image: geralt/Pixabay