Announcing GXL Sy

Today, we are launching Sy, a system for agent-native search and synthesis over biomedical preprints.

There are more papers, repositories, benchmarks, and discussions than any human team can track. Preprint volume continues to grow rapidly year over year, making it increasingly difficult for scientists to stay current, connect results across papers, and extract reliable conclusions from the literature. Sy helps researchers understand and query biomedical preprints at scale.

Sy is built for large-scale literature analysis. It enables researchers to run aggregation queries across many papers at once, reconcile findings across sources, and produce evidence-weighted conclusions with exact citations to the original documents. In internal benchmarks across full-text Q&A, idea novelty checking, and deep cross-paper synthesis, the system is 1.6× more accurate, 2.4× faster, and 3.6× cheaper than state-of-the-art MCP-based approaches.

Sy combines deep indexing, agent-native workflows, and multimodal analysis to make the scientific literature more computable:

• Deep indexing: Sy indexes full paper contents, including metadata, titles, abstracts, main text, and supplements.
• Agent-native: Sy works through a virtual filesystem that mirrors code environments, making it natural for agents to navigate and analyze research materials.
• Massively parallel analysis: Parallel agents can surface relevant information from hundreds of papers at a time, rather than being limited to a small fixed context.
• Multimodal: Sy can draw information from figures, tables, and data files—not just plain text.
• Accurate and traceable: Every response is grounded in source documents and traceable to the exact location where the evidence appears.

Sy especially shines on complex queries involving complex knowledge synthesis, such as:

• Drafting review-paper style syntheses across a body of literature
• Tracing how a particular paper has been used in subsequent work
• Identifying trends, disagreements, and emerging themes across hundreds of papers

Rather than helping researchers read one paper at a time, Sy is designed to help them reason over the literature as a whole. 

See here for more information about our approach.