Agents-K1: Towards Agent-native Knowledge Orchestration

Agents-K1 is a newly introduced end-to-end knowledge orchestration pipeline designed to enhance scientific reasoning by converting raw documents into agent-native scientific knowledge graphs. It features a multimodal parser that captures comprehensive data from entire papers, a 4B parameter information-extraction backbone trained with GRPO, and a tri-source agent interface for unified data retrieval. This framework processes 2.46 million scientific papers to create the Scholar-KG, with a one-million-paper subset released, demonstrating significant improvements in information extraction and multi-hop reasoning, which is crucial for practitioners developing AI systems that require deep scientific understanding.