Agents-K1: Towards Agent-native Knowledge Orchestration

2026-06-11

AI Summary

This work addresses the limitation of existing large language model–driven research agents, which typically rely on abstracts and flat citations while overlooking critical elements such as entities, claims, evidence, and methodological lineages that underpin scientific reasoning. To overcome this, the authors propose the first agent-native framework for scientific knowledge orchestration, featuring an end-to-end pipeline that enables full-text structured knowledge extraction and cross-document reasoning. The system integrates a five-module multimodal parser, a 4B-parameter information extraction backbone trained via GRPO reinforcement learning, and a unified graph-based retrieval interface. Built upon 2.46 million papers across six scientific domains, the resulting Scholar-KG knowledge graph—accompanied by a publicly released one-million-paper subset—demonstrates substantial improvements over current approaches in scientific information extraction, knowledge graph construction, and multi-hop reasoning tasks.

Abstract

Current LLM-based research agents have advanced through agent orchestration, yet largely overlook scientific knowledge orchestration. Existing works often reduce papers to abstracts, surface mentions, and flat \texttt{cites} edges, omitting key entities, claims, evidence, mechanisms, and method lineages essential for scientific reasoning. To this end, we introduce \textbf{Agents-K1}, an end-to-end knowledge orchestration pipeline that converts raw documents into agent-native scientific knowledge graphs. Agents-K1 integrates three components under a unifying theoretical foundation: a multimodal parser whose five-module schema captures entities, multimodal evidence, citations, and typed inter-entity relations across the full paper rather than abstracts alone; a 4B information-extraction backbone trained with GRPO under a rule-based reward; and a graphanything CLI, a tri-source agent interface that unifies web search, multimodal graph retrieval, and cross-document traversal. On top of this, we process 2.46 million scientific papers across six subjects to produce \textbf{Scholar-KG}, of which we release a one-million-paper subset, and the full Scholar-KG is accessible via the SCP link below. The same pipeline can be extended to general-domain corpora and to schema-conformant data synthesis. Extensive experiments demonstrate that Agents-K1 achieves superior performance in scientific information extraction, knowledge graph construction, and multi-hop scientific reasoning.

Problem

• scientific knowledge orchestration
• knowledge graph
• information extraction
• scientific reasoning
• agent-native

Innovation

• knowledge orchestration
• scientific knowledge graph
• multimodal parsing
• agent-native reasoning
• GRPO-based information extraction