Towards Retrieving Interaction Spaces for Agentic Search

2026-06-04

AI Summary

Traditional retrieval methods struggle to support efficient interactive exploration by agents within large-scale corpora. This work proposes constructing an “interaction space” by first using BM25 retrieval to identify a relevant document subset and then preprocessing documents during indexing to enable structured navigation via shell-like tools (e.g., grep, file reading). This approach shifts the retrieval objective from document ranking to building an actionable subspace tailored for agent interaction. On BrowseComp-Plus, it achieves 78% of the accuracy of a pure shell-based method while using only about one-quarter of the query budget; at a scale of one million documents, it attains 81% accuracy, significantly outperforming baseline methods whose performance degrades under such conditions.

Abstract

Retrieval for search agents is still inherited from non-agentic information retrieval: a retriever ranks the corpus and the agent reads a small set of returned documents. Recent direct corpus interaction (DCI) work shows that agents can instead interact with the raw corpus through shell tools such as grep and file reads. But unbounded interaction does not scale: every broad shell command is a scan over the whole corpus, and latency degrades sharply as the corpus grows. We argue that the role of retrieval for agentic search is not just to select documents that fit in the LLM context window, but to construct an interaction space: a bounded subset of the corpus the agent can explore with associated tools. Two design consequences follow. The space needs a boundary supplied by retrieval, and the objects within it should be processed for interaction. As a proof of concept, we propose RISE (Retrieving Interaction SpacE): we use BM25 to construct the interaction space; meanwhile, its documents are processed during indexing for shell-style navigation. On BrowseComp-Plus, RISE matches the pure-shell DCI baseline at 78% accuracy with gpt-5.4-mini at roughly one quarter of the per-query cost. At 1M documents, RISE-BM25 reaches 81% on gpt-5.4-mini, whereas DCI on gpt-5.4-nano degrades to 60% with 33 of 100 wall-clock failures.

Problem

• agentic search
• retrieval
• interaction space
• scalability
• direct corpus interaction

Innovation

• agentic search
• interaction space
• direct corpus interaction
• retrieval
• RISE