Nace AI announced the release of Drex for Search, a new retrieval layer aimed at optimizing how human operators and AI agents search across large document sets. The system interfaces directly with agentic coding environments such as Claude Code, Cursor, and OpenCode via CLI, SDKs, MCP, or REST endpoints. According to the company, Drex frames search as a sequence of decision steps, evaluating options with calibrated probabilities in a single forward pass to lower operational overhead compared to traditional fixed-depth RAG systems.
In benchmark evaluations reported by Nace AI, Drex for Search achieved a 92.05% score on BrowseComp-Plus using GPT-5 with an average of 12.6 search calls per query. On the OfficeQA Pro benchmark across U.S. Treasury Bulletins, Nace reported 6% higher accuracy, 62% fewer steps, and 56% fewer tokens consumed. Furthermore, on EnterpriseRAG-Bench involving over 500,000 enterprise documents, the platform reached an overall score of 88.22, outperforming OpenAI File Search.
Drex for Search is accessible through the Nace API Console. Developers and enterprise operators can integrate the capability using a single API key across four access points.
Why it matters
Replaces traditional top-k RAG chunking with probabilistic decision passes, drastically cutting token usage and latency for complex multi-hop searches.
Delivers seamless integration with major developer tools like Cursor and Claude Code via standardized MCP, CLI, and REST protocols.
Demonstrates significant enterprise search benchmark gains over built-in options like OpenAI File Search across large internal document repositories.
Source: marktechpost.com



