Dr-DCI aims to scale agentic search over large document collections

A new arXiv paper introduces Dr-DCI, a framework for scaling direct corpus interaction in agentic search. The method combines retriever-mediated discovery with shell-executable corpus operations for filtering, comparison, and verification.

The authors argue that ordinary retrieval interfaces expose evidence as ranked results or bounded document views, which limits an agent’s ability to reorganize material and check constraints. Direct corpus operations are more flexible but can become slow at scale.

Dr-DCI is designed to balance those tradeoffs, making large-corpus research agents more efficient and verifiable.