Nexthink deploys OpenSearch vector search for Spark agent

Per a guest post on the AWS Blog authored by Nexthink, the company implemented Amazon OpenSearch Service vector search together with Amazon Bedrock embeddings and infrastructure-as-code to support its autonomous IT-support agent Spark. The blog reports Spark resolves issues at a 77% first-contact resolution rate and is deployed across 12 AWS Regions. The post says embeddings were generated using V2 through Amazon Bedrock and that retrieval must operate at sub-second latency across millions of documents to serve enterprise users.

Per the AWS blog post, Nexthink built a retrieval layer on OpenSearch Service with vector search to enable semantic matching of user queries to documentation, historical resolutions, and remediation scripts. The post emphasises semantic retrieval over keyword matching and cites the operational risk of incorrect retrieval leading to dangerous automated commands as a guardrail rationale for high-precision vectors and verification before action.

Industry-pattern observations: production AI agents increasingly rely on vector search plus high-quality embeddings to bridge natural-language queries and operational knowledge bases. Practitioners deploying autonomous remediation should expect engineering effort around embedding quality, indexing strategies, latency optimization, and safe execution gating.

For practitioners: monitor retrieval latency at scale, embedding drift over time, and mechanisms the team uses for verification before action. Public metrics such as first-contact resolution and regional deployments provide useful operational benchmarks when evaluating similar agent architectures.