Orca Security Launches Discovery AI for Natural Language Security Queries Using Amazon Bedrock
Orca Security has introduced Discovery AI, a system enabling security professionals to query cloud infrastructure using plain English. Built on Amazon Bedrock, it converts natural language into Sonar queries with high efficiency and accuracy.
What changed?
Orca Security released Discovery AI, a tool that allows users to express complex security queries in natural language. Leveraging Amazon Bedrock, Discovery AI translates these questions into valid Sonar queries without requiring the user to understand the schema or syntax of Orca's domain-specific language. The system operates across more than 900 cloud models and achieves major performance improvements: 90% lower query costs, 93% fewer tokens used, and 75% lower response latency than the initial implementation. The architecture uses a nine-stage pipeline combining retrieval-augmented generation, multi-model orchestration, and schema-aware query construction.
Why does it matter to an everyday developer?
Developers and security analysts no longer need to invest time in learning Orca’s complex query schemas or operator syntax to extract insights from cloud environments. The natural language interface democratizes access to security data and reduces the expertise barrier, allowing teams to formulate sophisticated queries quickly. Lower costs and improved query performance also enable more frequent and interactive use, which may improve security review cycles and reduce response times to incidents. Support for over 900 cloud infrastructure and application models gives broad coverage, while the pipeline’s continuous improvement mechanisms help the system adapt to evolving cloud environments.
What can the developer do now?
If you use Orca Security, you can now issue natural language queries to inspect cloud resources, review compliance, or investigate incidents—without writing Sonar queries manually. Security and DevOps teams can experiment with diverse question phrasing, as the system's agentic, multi-stage pipeline is designed to robustly interpret varied inputs. You do not need to update integrations or migrate API usage, as the change is additive and does not introduce breaking changes. Domain experts can further participate in fine-tuning results via the internal labeling tool, which provides feedback for continuous improvement of query quality.
