Reproduce “DeepXiv-SDK: An Agentic Data Interface for Scientific Literature” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hongjin Qian, Ziyi Xia, Ze Liu, Jianlyu Chen, Kun Luo, Minghao Qin, Chaofan Li, Lei Xiong and 6 more
DeepXiv-SDK provides a structured, agentic data interface for scientific literature, transforming raw papers into tool-callable JSON objects that support progressive access and budget hints. This system enhances LLM agent efficiency and output quality in scientific search and deep research question answering through normalized, section-addressable information, demonstrating faster retrieval and reduced token usage compared to traditional methods.
Key insights
- Treating scientific paper access as a structured, deterministic data interface rather than an isolated parsing step significantly improves LLM agent performance and efficiency.
- Progressive disclosure of information (coarse-to-fine views) is crucial for enabling agents to efficiently screen candidates and selectively engage with relevant content, optimizing token expenditure.
- Embedding budget hints and semantic signals (e.g., TL;DRs) directly into the data interface guides agents towards more cost-aware and evidence-seeking interactions with scientific literature.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.