As AI agents increasingly navigate the web, their need for structured, accessible knowledge grows—especially in the life sciences where agentic pipelines are proliferating. Traditional literature resources like Europe PMC, which indexes over 40 million records, were designed for human users: they expect keyword queries and return full papers, forcing agents to learn complex syntax, run multiple searches, and read entire documents to find relevant evidence.
To address this, researchers at EMBL and Hugging Face introduce EMBL AI Librarian, a knowledge layer that upgrades the Europe PMC interface for AI agents. Instead of wrestling with syntax, an agent simply asks a question in natural language and receives evidence snippets that directly answer it. A single large language model (LLM) orchestrates the entire retrieval process: it plans complementary subqueries, executes them via the live Europe PMC search engine, then reads the selected papers to pinpoint the relevant evidence.
The system was evaluated across four benchmarks: literature synthesis, claim verification, open-domain question answering, and downstream biology tasks such as protocol questions and sequence manipulation. On ScholarQABench, Librarian improved Citation F1 by more than 16 points over strong recently published baselines. When used as the retrieval layer in an existing claim-verification pipeline, it increased agreement with expert consensus. On the open-form LitQA2 benchmark, a GPT-5.4 agent scored about 8 points higher when grounded in Librarian than with web search.
The authors highlight that Librarian requires no vector database or separate literature index, and it remains always up to date by querying the live Europe PMC. The code is publicly available on GitHub.