RARG streamlines ripgrep search for AI agents

Direct corpus interaction tools like ripgrep are essential for LLM agents, but unguided file traversal wastes tokens and slows down convergence.

• –Converts document relevance into an execution prior to direct corpus interaction rather than relying solely on naive keyword traversal.
• –Employs a coarse-to-fine guidance mechanism combining sequential document ordering, query-focused entry points, and match-level reranking.
• –Outperforms both standard retrieval-augmented generation (RAG) baselines and unguided direct-interaction agents in accuracy and search speed.