The search engine, the directory, and the marketplace for all of life science — built for the humans, and the AIs, now doing the science.
35M+ papers, 250M+ proteins, 226K+ structures and millions of datasets sit in dozens of silos with no shared index, no map, and no search across them.
The window to become the default index — before each lab's in-house graph calcifies into permanence — is open now.
doi.bio is the search engine, the directory, the marketplace, and the ad network for life science — over a pre-connected graph stored as the plain markdown that agents read natively.
An instant, keyboard-first omnibox over our own index — the speed of "Google for gamers," pointed at all of biology. Slash commands route anywhere; everything else searches the graph live.
Every entity gets a clean, addressable home page — markdown for humans, an API for machines — cross-linked into the graph. Browsable by a person, crawlable by an agent.
Wikipedia's contributor flywheel and GitHub's openness — with a paycheck. We never hold the original: you keep your data, we hold the map, and the money routes back to you.
ORCID ↔ repo handshake auto-verifies you. No server to run, no auth backend.you.doi.bio — searchable, linked into the graph, machine-readable.70%.The open baseline (PubMed, PDB, UniProt) pulls users in. Premium datasets and pay-per-use monetize them — and doi.bio takes 20–30% of every transaction.
Every page is markdown an agent reads natively and an endpoint it can pay. Agents buy exactly the data a model needs — per query, per token, per dataset — and the money routes to the scientist who made it.
Every page already knows its exact biological context — the protein, the disease, the lab. That's the most precisely targeted ad inventory in science. Advertisers: pharma, reagent & instrument vendors, CROs, recruiters.
Steven Ness — PhD, Computer Science. Building doi.bio solo: ingestion, entity resolution, indexing, and a 7.27M-article search engine.
Re-ran the targets through current AF3-class predictors — chai-1, Protenix, Boltz, IntelliFold, Gnina. Best predicted pose per target; the field's "success" bar is ≈ 2 Å RMSD.
S-adenosylhomocysteine (SAH) — predicted pose 0.14 Å RMSD, LDDT-PLI 0.97. Protenix · T1124.
A drug-like inhibitor (DW0) — 0.20 Å, LDDT-PLI 0.99. Boltz · T1188.
An N-glycan (NAG) — 0.34 Å, LDDT-PLI 0.96. Boltz · T1187.
A lean, solo-first round — ~18 months to ship the wedge (search over the live graph + the first paid datasets) and reach first revenue. Then raise a seed from strength.
On a SAFE — default $250K, room to expand to ~$400K for the right partner.