Ariadne
From English prompt to executed workflow.
State a goal in plain English — “design an enzyme that does X at Y conditions” — and Ariadne handles literature search, workflow composition, submission to compute, monitoring, and analysis of the results, all referenceable in one conversation.
What it does
Ariadne is a copilot for protein design. You describe the outcome you want; it composes and runs the workflow. Behind a single conversation it searches the literature, assembles the right pipeline, submits it to compute, tracks progress, and helps you interpret the results — so scientists spend their time on the science, not on plumbing.
The literature engine
Ariadne rides on a multi-tier literature-discovery engine. Rather than keyword hits, it works through a retrieval cascade — from a broad survey down to curated, workflow-level structure — so the copilot reasons over what has actually been done in a field, and grounds each design decision in prior work.
›
Tier 1 — Broad survey — wide recall across the literature (~40 million abstracts)
›
Tier 2 — Focused retrieval — topic- and method-scoped passages (~4 million fulltext)
›
Tier 3 — Curated structure — entities, relations, and protocols extracted (~14,000 curated protein engineering papers)
›
Tier 4 — Workflow-level — prior pipelines mapped to reusable steps (~14,000 papers and ~300 tools)
Workflow memory
Every workflow Ariadne runs is given a memorable, Docker-style name. Refer back to it later in chat and Ariadne pulls it from the backend — its parameters, inputs, and results — so experiments are recallable and composable rather than one-off.