Paper 2605.05242
Beyond Semantic Similarity: Rethinking Retrieval for Agentic Search via Direct Corpus Interaction
- Published
- May 2026
- Research lab
- Independent
- Citations
- 21
- GitHub
- 382 stars
01 In brief
Summary
The paper introduces Direct Corpus Interaction (DCI), a retrieval paradigm where agents search raw corpora using terminal tools (grep, bash, file reads) instead of conventional retrievers.
DCI bypasses embedding models and vector indexes, enabling fine-grained, compositional search.
Evaluated on BrowseComp-Plus, multi-hop QA, and IR benchmarks, DCI-Agent-CC (Claude Sonnet 4.6) achieves 80.0% accuracy on BrowseComp-Plus (vs.
69.0% for retrieval agents) with 29.4% lower cost, 83.0% average accuracy on QA (vs.
52.3% for ASearcher-Local-14B), and 68.5 NDCG@10 on IR (vs.
47.0% for ReasonRank-32B).
DCI-Agent-Lite (GPT-5.4 nano) also outperforms baselines.
Ablations show DCI's gains stem from 'retrieval interface resolution'—converting surfaced evidence into local verification steps—rather than higher recall.
DCI scales well in depth but costs rise with corpus breadth.
Context management (truncation, compaction, summarization) is crucial, with L3 policy optimal.
The paper reframes retrieval as an interface-design problem, not just retriever design, for capable agents.
02 From the paper
Abstract
Modern retrieval systems, whether lexical or semantic, expose a corpus through a fixed similarity interface that compresses access into a single top-k retrieval step before reasoning. This abstraction is efficient, but for agentic search, it becomes a bottleneck: exact lexical constraints, sparse clue conjunctions, local context checks, and multi-step hypothesis refinement are difficult to implement by calling a conventional off-the-shelf retriever, and evidence filtered out early cannot be recovered by stronger downstream reasoning. Agentic tasks further exacerbate this limitation because they require agents to orchestrate multiple steps, including discovering intermediate entities, combining weak clues, and revising the plan after observing partial evidence. To tackle the limitation, we study direct corpus interaction (DCI), where an agent searches the raw corpus directly with general-purpose terminal tools (e.g., grep, file reads, shell commands, lightweight scripts), without any embedding model, vector index, or retrieval API. This approach requires no offline indexing and adapts naturally to evolving local corpora. Across IR benchmarks and end-to-end agentic search tasks, this simple setup substantially outperforms strong sparse, dense, and reranking baselines on several BRIGHT and BEIR datasets, and attains strong accuracy on BrowseComp-Plus and multi-hop QA without relying on any conventional semantic retriever. Our results indicate that as language agents become stronger, retrieval quality depends not only on reasoning ability but also on the resolution of the interface through which the model interacts with the corpus, with which DCI opens a broader interface-design space for agentic search.