A Tale of Two Graphs: Separating Knowledge Exploration from Outline Structure for Open-Ended Deep Research

Open-Ended Deep Research (OEDR) pushes LLM agents beyond short-form QA toward long-horizon workflows that iteratively search, connect, and synthesize evidence into structured reports. However, existing OEDR agents largely follow either linear "search-then-generate" accumulation or outline-centric planning. The former suffers from lost-in-the-middle failures as evidence grows, while the latter relies on the LLM to implicitly infer knowledge gaps from the outline alone, providing weak supervision for identifying missing relations and triggering targeted exploration. We present DualGraph memory, an architecture that separates what the agent knows from how it writes. DualGraph maintains two co-evolving graphs: an Outline Graph (OG), and a Knowledge Graph (KG), a semantic memory that stores fine-grained knowledge units, including core entities, concepts, and their relations. By analyzing the KG topology together with structural signals from the OG, DualGraph generates targeted search queries, enabling more efficient and comprehensive iterative knowledge-driven exploration and refinement. Across four established OEDR benchmarks, DualGraph consistently outperforms state-of-the-art baselines in report depth, breadth, and factual grounding; for example, it reaches a 53.08 RACE score on DeepResearch Bench with GPT-5. Moreover, ablation studies confirm the central role of the dual-graph design.

People increasingly use AI systems to research broad questions and write structured reports, but these systems often lose track of what they have already found or fail to notice what is still missing. Many current systems either collect search results in a long sequence or follow a report outline that may hide important knowledge gaps. We introduce DualGraph, a memory design that helps an AI research agent separate two tasks: understanding the topic and organizing the final report. It keeps one graph for the report outline and another graph for the facts, entities, concepts, and relationships discovered during search. By comparing these two graphs, the agent can decide what to search for next, instead of only asking broad follow-up questions. This makes the research process more targeted, helps fill missing connections, and reduces the chance of shallow or poorly supported reports. In tests on several deep-research benchmarks, DualGraph produced reports that were broader, deeper, and better grounded in evidence than strong existing systems. Our results suggest that better memory structures can make AI research assistants more reliable for long, open-ended investigations.