Tag
RouteDRAG 是一个基于学习的框架，用于在多模态文档 RAG 中决定是否调用 Agent 及其执行路径，在两个基准上比仅用短路径执行平均提升 3.47 个百分点，且仅将 79.3% 的查询分配给 Agent。
This paper investigates agentic retrieval that combines LLM reasoning with dense retrievers in a ReAct loop for complex search tasks, showing an 8.7-point nDCG@10 improvement over standard retrieval and strong generalization on ViDoRe v3 and BRIGHT leaderboards, but at a substantial latency and token cost of 107.4 seconds and 764K input tokens per query.
PatchHolmes, accepted at AACL-IJCNLP 2026, uses an agentic listwise selection approach to pair software vulnerabilities with their fixing commits, achieving 59.95% Recall@1 versus 34.61% for pointwise baselines by having the agent inspect the full candidate list and reading ~96k tokens per query on a frozen open-weight model.
CoVeR is a coverage-based routing method that reduces LLM verifier calls by 62-68% in agentic retrieval systems while maintaining accuracy on multi-hop QA benchmarks.
TRACE is a training-free agentic retrieval framework designed for accountable source discovery in historical digital archives, outperforming RAG baselines on a benchmark with low cost per question.
InsightEmb is a contrastive embedding framework for agentic insight retrieval that learns progress-oriented retrieval geometry from mathematical reasoning data alone, improving retrieval for LLM agents without environment-specific training.
LlamaParse introduces new document retrieval endpoints including hybrid search, file grep, file find, and file read, aiming to improve agentic retrieval quality over unstructured documents.
Jerry Liu highlights the engineering challenges of productionizing agentic retrieval systems, emphasizing that success depends on careful tuning of chunking, synchronization, reranking, and tool API design rather than novel techniques.
NVIDIA releases Nemotron 3 Embed, a collection of open embedding models that top the RTEB leaderboard, featuring an 8B flagship model and efficient 1B variants for production-scale retrieval.
LlamaIndex has created a Retrieval Harness for modern agentic retrieval, providing a persistent data pipeline for connecting, indexing, and querying large knowledge bases with tools like semantic search and regex grep, allowing agents to autonomously navigate knowledge bases.
A blog post revisits the concept of 'Slow Search' in the context of agentic retrieval, arguing that per-query latency can be traded for better retrieval quality to reduce overall task time and cost for AI agents.
Introduces Infini Memory, a maintainable text-based persistent memory architecture for LLM agents that uses topic-structured documents and iterative retrieval to improve long-term memory usage, achieving 64.7% on MemoryAgentBench.
MemDreamer decouples perception and reasoning for long video understanding using hierarchical graph memory and agentic retrieval, achieving state-of-the-art performance with reduced computational overhead.
This paper introduces a fleet of five sensor-specialized Mini-JEPA foundation models for hydrologic intelligence, achieving high reconstruction accuracy (R² up to 0.97) and outperforming the Google AlphaEarth generalist on physics-matched tasks when routed via an LLM agent.