Multimodal large language models (MLLMs) have shown strong visual reasoning abilities, but knowledge-intensive visual question answering often requires external textual evidence beyond the image and the model’s parametric knowledge. Existing multimodal RAG systems commonly rely on Top- retrieval or reranking, which may return redundant passages and provide limited control over whether an answer update is sufficiently supported by the retrieved evidence. We propose CLIMB, a training-free inference-time framework for multimodal RAG. CLIMB first constructs a compact complementary evidence pool using an MMR-style objective that balances query relevance and passage-level redundancy. It then performs confidence-controlled refinement within this fixed pool: an R/E/C critic scores passages by relevance, evidence specificity, and cross-modal alignment, while an evidence-grounded confidence estimator accepts an updated answer only when the estimated confidence increases. This design provides a simple stopping criterion and reduces unnecessary refinement without modifying the underlying retriever or MLLM. Experiments on Encyclopedic-VQA and InfoSeek show that CLIMB consistently improves over retrieval-augmented multimodal baselines. Ablations further indicate that complementary pooling, critic-based scoring, and iterative confidence-controlled refinement each contribute to the final performance.

Graph-structured retrieval-augmented generation (GraphRAG) has shown strong potential for improving multi-hop reasoning in large language models. However, existing approaches typically depend on dense entity–entity linking, which is computationally expensive, susceptible to hallucinated relations, and difficult to maintain under continual updates. We present Bipartite Graph Retrieval-Augmented Generation (BG-RAG), a lightweight bipartite framework that organizes knowledge into three layers, from chunks to entities to concepts, and replaces direct entity links with concept-mediated connections. This design enables more efficient graph construction, reduces reliance on hallucination-prone relation-edge prediction by avoiding explicit entity–entity relation extraction, and naturally supports incremental updates without global reconstruction. Experiments on HotpotQA, 2Wiki, and MuSiQue show that BG-RAG achieves competitive performance against RAG and GraphRAG baselines. Its concept-mediated structure also provides a lightweight alternative to dense entity–entity graph construction. These results suggest that BG-RAG is a lightweight alternative for retrieval-augmented generation, particularly when avoiding dense entity–entity linking is desirable.

While Large Language Model (LLM) based agents excel at complex tasks, their performance in open-ended scenarios is often constrained by isolated operation and reliance on static databases, missing the dynamic knowledge exchange of human dialogue. To bridge this gap, we propose the INteractive Memory Sharing (INMS) framework, an asynchronous interaction paradigm for multi-agent systems. By integrating real-time memory filtering, storage, and retrieval, INMS establishes a shared conversational memory pool. This enables continuous, dialogue-like memory sharing among agents, promoting collective self-enhancement and dynamically refining the retrieval mediator based on interaction history. Extensive experiments across three datasets demonstrate that INMS improves agent performance by effectively modeling multi-agent interaction and collective knowledge sharing.

By augmenting Large Language Models (LLMs) with external tools, their capacity to solve complex problems has been significantly enhanced. However, despite ongoing advancements in the parsing capabilities of LLMs, incorporating all available tools simultaneously in the prompt remains impractical due to the vast number of external tools. Consequently, it is essential to provide LLMs with a precise set of tools tailored to the specific task, considering both quantity and quality. Current tool retrieval methods primarily focus on refining the ranking list of tools and directly packaging a fixed number of top-ranked tools as the tool set. However, these approaches often fail to equip LLMs with the optimal set of tools prior to execution, since the optimal number of tools for different tasks could be different, resulting in inefficiencies such as redundant or unsuitable tools, which impede immediate access to the most relevant tools. This paper addresses the challenge of recommending precise toolsets for LLMs. We introduce the problem of tool recommendation, define its scope, and propose a novel Precision-driven Tool Recommendation (PTR) approach. PTR captures an initial, concise set of tools by leveraging historical tool bundle usage and dynamically adjusts the tool set by performing tool matching, culminating in a multi-view-based tool addition. Additionally, we present a new dataset, RecTools, and a metric, TRACC, designed to evaluate the effectiveness of tool recommendation for LLMs. We further validate our design choices through comprehensive experiments, demonstrating promising accuracy across two open benchmarks and our RecTools dataset.