Knows introduces structured research format for AI agents to access academic papers
There's a new standard called Knows aimed at improving how AI agents handle academic research. Research is usually stored as PDFs, which creates challenges for large language models when they try to pull out specific details from lengthy texts. This process can be inefficient and inconsistent, especially at scale. Knows introduces a streamlined specification that adds structured claims, evidence, and verifiable links directly into research artifacts, allowing LLM agents to use them easily. It includes a lightweight YAML file called KnowsRecord that sits alongside the original PDFs without changing them, and it’s checked with a schema linter. In testing, 140 questions were asked across 20 papers from 14 fields, evaluating different setups with six LLM agents.
Key facts
- Knows is a lightweight companion specification for research artifacts
- Research artifacts are primarily distributed as PDF documents
- PDFs create bottlenecks for agent-assisted research workflows
- LLM agents need to infer fine-grained information from lengthy documents
- Knows binds structured claims, evidence, provenance, and verifiable relations to research
- Uses a thin YAML sidecar called KnowsRecord that coexists with original PDFs
- Requires no changes to the original publication
- Evaluated on 140 comprehension questions across 20 papers spanning 14 disciplines
- Compared PDF-only, sidecar-only, and hybrid conditions across six LLM agents
Entities
—