.
Every model, tool and dataset in one place.
Fold a protein, pull any of 247M structures, search the literature — no GPU, no account, no credit card.
What you get
Three things, and they work together.
Most of computational biology is spent getting tools to run at all. This is the part where you skip that.
- 01
Ask in plain language
Describe what you want to know. It finds the right models, runs them, queries the data, reads the papers, and shows you the evidence behind every number.
Which of these twelve inhibitors is most likely to bind KRAS G12C, and is there any published affinity data?Ran Boltz-2 on 12 ligands · queried ChEMBL · read 3 papersThree candidates dock into the switch-II pocket with a predicted affinity below 100 nM. Compound 7 is the strongest, and ChEMBL reports a measured IC50 of 38 nM for it.Compound 738 nMCompound 271 nMCompound 1194 nMBoltz-2ChEMBL 34Ostrem et al. 2013PDB 6OIM - 02
Run any model, instantly
Folding, docking, protein design, genomics, chemistry. Paste your input and get a result — no GPU to rent, no environment to build, no weights to download.
ESMFoldProtein structure · 3B paramsFinished in 4.2 s>sp|P69905|HBA_HUMANMVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFLSFPTTKTYFPHFDLSHGSAQVKGHGKKVADALTNAVAHVDDMPNALSALSDLHAHKLMean pLDDT88.6Very highConfidentLow - 03
Search inside the data
Not just dataset titles — the rows. Filter, sort and query hundreds of scientific datasets and live databases without downloading a single file.
ClinVargene = BRCA1significance ∈ pathogenic2,184 rowsVariant Consequence Significance Review c.5266dupC Frameshift Pathogenic ★★★★ c.68_69delAG Frameshift Pathogenic ★★★★ c.181T>G Missense Pathogenic ★★★ c.4327C>T Nonsense Pathogenic ★★★ c.5123C>A Missense Likely pathogenic ★★ Live from the source · no downloadSELECT * FROM clinvar WHERE gene = 'BRCA1' …
Why we built it
Built by the people who kept hitting this wall.
The best models in biology are open and free, and almost nobody can run them. We spent years on both sides of that gap — writing the models, and waiting on the cluster to use them. So we closed it.
What's inside
Every model, every dataset.
One place to run them.
Protein structure, drug design, genomics, chemistry, imaging and simulation — across every scientific field, all reachable the same way.
Models
and 248 datasets, catalogued and searchable.
Free, and instant
Fold a protein, pull a predicted structure, look up a compound, search the literature. No account, no card, no waiting.
141
Everything else, on demand
The rest of the library runs on our servers the moment you ask. Pay for the seconds it takes, nothing while it sits idle.
The data, already connected
44 live databases queried directly. No accounts, no formats to convert, and every record carries where it came from.
- ReactomePathways and reactions
- Harvard DataverseResearch datasets
- Zenodo4M+ records
- Figshare10M+ items
- Gene Expression Omnibus250k+ series
- DANDI Archive1k+ dandisets
Across every field
Protein structure, drug design, genomics, chemistry, imaging and simulation, all reachable the same way.
Work you can defend
A result you can't check is a result you can't publish.
Science has a higher bar than a plausible-sounding answer, and everything here is built for the moment a reviewer asks where a number came from.
- 01
Every claim is traceable
Each number links back to the model that computed it or the paper it came from. Nothing is asserted from memory, and nothing is dressed up when the data is thin.
Compound 7 docks into the switch-II pocket with a predicted affinity of 38 nM1, consistent with the measured IC50 reported for ARS-16202.
1 · Computedrun a3f9c2Boltz-2 · v2.1.112 ligands, 5 seeds, 200 recycles2026-09-04 · 41 s on 1 GPU2 · Publisheddoi:10.1016/j.cell.2018.01.006Janes et al., Cell 2018Targeting KRAS mutant cancers with a covalent G12C-specific inhibitor - 02
Every run is reproducible
Inputs, model version, parameters and outputs are kept together. Re-run it in a year and you will know exactly what produced the original.
run 7d21f4 · 2025-09-02Re-run- Input
- sha256 · 9b1e…4c07 (1 sequence, 141 aa)
- Model
- ESMFold · v1.0.3 · weights 2a77…e0f1
- Parameters
- num_recycles = 3 · chunk_size = 128
- Output
- PDB · 141 residues · mean pLDDT 88.6
Re-run today · identical output, byte for byte - 03
Your data stays yours
Sequences, structures and compounds you submit are used to answer your question and for nothing else. Never training data — not ours, not anyone's.
Your submissionprivate>candidate_07 · SMILES · CC(C)N1C(=O)N(C)c2ccc(Cl)cc2C1=O…- Answering this question
- Your own run history
- Training any model
- Shared with anyone else
Ask it something today.
Bring a sequence, a compound, a paper or a question. You will have an answer before you have finished setting up the alternative.
No credit card. No setup. Nothing to install.