Imagine asking an AI to summarize the latest breakthroughs in battery technology, only to have it hallucinate a connection that never existed. For scientists, this is a daily nightmare. Current search tools return long lists of papers, forcing researchers to manually hunt for specific facts, verify their origins, and stitch together answers from disparate sources. It is a tedious, error-prone process that slows down discovery and risks spreading misinformation.

Researchers have introduced AskChem to fix this broken workflow by changing the fundamental unit of search. Instead of retrieving entire documents, AskChem breaks chemistry literature down into atomic, typed claims. Each claim is grounded by a specific source DOI and a verbatim quote or precise evidence locator. This creates a shared claim store that allows for hierarchical browsing through a stabilized faceted taxonomy and links claims together in an evidence graph. The system currently indexes 2.4 million claims from 147,000 papers, offering both a web interface and direct access for AI agents via REST, SDK, and MCP protocols.

The results are striking. When researchers grounded a GPT-5.5 reader in AskChem, the system achieved a 100% resolvable DOI rate, compared to just 88.3% without retrieval. It also demonstrated the highest citation density among five tested systems. This means AI agents can now synthesize cross-paper answers with unprecedented accuracy, eliminating the guesswork and manual verification that currently plague chemical research.

AskChem represents a shift from document-centric search to claim-centric infrastructure. By making every fact traceable to its source, it empowers both human scientists and AI agents to build knowledge on a solid foundation. The platform is now live, offering a new standard for how we interact with scientific literature.