Orca Security launches Discovery AI on Amazon Bedrock

Orca Security and AWS have released Discovery AI, a natural‑language query agent built on Amazon Bedrock. The service translates plain‑text questions into valid Sonar filters, allowing analysts to search across Orca’s catalog of more than 900 cloud infrastructure models without writing JSON queries.

Discovery AI uses a nine‑stage pipeline that combines retrieval‑augmented generation with schema‑aware construction. Early stages include a deterministic matcher for structured inputs and a guardrail that blocks unsafe requests. Semantic retrieval pulls the 20 nearest‑neighbor examples using Amazon Titan embeddings, which Cohere Rerank 3.5 then narrows to five. Anthropic Claude Haiku 4.5 drives model selection and generation, while Amazon Nova 2 Lite handles validation. Orca reports a 90 % reduction in query cost, a 93 % drop in input tokens, and a 75 % cut in latency compared with the initial baseline.

The architecture demonstrates how multi‑model access via Bedrock, example‑based prompting, and continuous evaluation can be combined to build production‑grade, cost‑effective AI tools for security analysts.