Context
Researchers introduce Direct Corpus Interaction (DCI), allowing agents to search raw corpora using general-purpose file and terminal operations rather than fixed sparse, dense or vector-search APIs. They report strong performance on selected BRIGHT, BEIR, BrowseComp-Plus and multi-hop QA evaluations.
What changed
The paper tests an agentic retrieval interface using grep, file reads and lightweight scripts without requiring an embedding model, vector index or offline indexing, arguing that direct corpus inspection can overcome early top-k filtering bottlenecks.
Why it matters
AI evidence discovery may depend on an agent's ability to inspect raw documents, combine constraints and revisit candidate sources, not only on similarity-ranking scores. This adds a distinct interface design to retrieval and source-selection research.
Visibility OS interpretation
Visibility OS interpretation: fine-grained corpus access could alter which documents are discovered and verified by agentic systems. The evidence does not establish adoption by commercial AI search engines or a universal advantage over indexed retrieval.
What remains uncertain
Does direct corpus interaction scale efficiently to live public-web search with access limits and dynamic pages? How does it compare on total cost, latency and security? Would it change citation distributions or publisher visibility in production?