Licences & Attribution

Running a model through Corollary does not grant you rights its licence withholds. The licence that governs your use is the one the model's authors published, not ours, and it travels with the output.

Corollary is itself a non-commercial demonstration, and its own restriction does not loosen a stricter upstream one. Where the two differ, the stricter applies.

Before you rely on any output: confirm the licence permits your field of use, comply with any attribution or share-alike condition, and cite the underlying model, dataset and publication — not Corollary alone.

Each runnable model carries one of three markers, shown on its model page. Of the 166 runnable entries in the catalogue today:

• Permitted (51) — a standard permissive licence such as MIT, Apache-2.0 or BSD-3-Clause. Commercial use is allowed by the upstream licence, subject to its attribution terms.
• Verify (110) — the licence has not been individually verified by us, or the model card states bespoke terms. Read the model card before you use it for anything. This is the largest group, and "verify" means exactly that: we have not checked it for you.
• Academic-only (5) — the licence forbids commercial use outright.

These markers are a navigation aid, not legal advice, and they can be wrong or out of date — an upstream repository can relicense at any time. The model card is authoritative; this catalogue is not.

These may not be used for commercial work under any circumstances:

• RoseTTAFold All-Atom (Baker Lab / UW) — MIT code / non-commercial weights. Code is MIT but the released weights are non-commercial. Commercial tier routes to Boltz-2 instead.
• RFDiffusion (Baker Lab / Institute for Protein Design) — Non-commercial (BSD-derived). Academic tier only. Commercial users are routed to BindCraft + ProteinMPNN for binder design.
• IgFold (Ruffolo et al., Johns Hopkins) — JHU Academic License
• Nucleotide Transformer v2 (InstaDeep / Wellcome Sanger) — CC-BY-NC-SA-4.0
• Galactica (Meta AI) — CC-BY-NC-4.0. Evaluation baseline only. Not routed for user-facing generation.

Beyond these, a substantial number of indexed models and datasets carry non-commercial Creative Commons terms (CC-BY-NC, CC-BY-NC-SA, CC-BY-NC-ND) or bespoke non-commercial licences from their authors. CC-BY-NC-ND additionally forbids distributing modified versions, which includes fine-tuned derivatives.

Some upstream repositories require you to accept terms or request access before the weights are released. Corollary lists such models and datasets so you can find them, but never runs them and never serves their rows: the platform does not accept terms on your behalf, and an acceptance given by us would not extend to you.

If you want to use a gated model or dataset, obtain access on your own Hugging Face account and accept the terms yourself, then run it in your own environment. The same applies to repositories that publish no licence at all — the platform does not execute or redistribute them, because silence is not permission.

A number of indexed models and datasets are under GPL-3.0, AGPL-3.0, LGPL, CC-BY-SA or ODbL. These impose obligations on what you distribute afterwards — typically that derivative works carry the same licence, and in the AGPL case that network use triggers a source-provision duty.

If you build on one of these, the obligation is yours. Check before you publish.

Corollary queries the following databases live, server-side, on your behalf. Each has its own terms of use, and several require attribution or restrict redistribution:

• Reactome — https://reactome.org
• Harvard Dataverse — https://dataverse.harvard.edu
• Zenodo — https://zenodo.org
• Figshare — https://figshare.com
• Gene Expression Omnibus — https://www.ncbi.nlm.nih.gov/geo/
• DANDI Archive — https://dandiarchive.org
• ICPSR — https://www.icpsr.umich.edu
• NACDA — https://www.icpsr.umich.edu/web/NACDA/index
• UniProtKB — https://www.uniprot.org
• PDB (RCSB) — https://www.rcsb.org
• AlphaFold DB — https://alphafold.ebi.ac.uk
• ChEMBL — https://www.ebi.ac.uk/chembl
• Europe PMC — https://europepmc.org
• ESM Atlas — https://esmatlas.com
• STRING — https://string-db.org
• InterPro — https://www.ebi.ac.uk/interpro
• BindingDB — https://www.bindingdb.org
• MGnify Human Gut (UHGG) — https://www.ebi.ac.uk/metagenomics/browse/genomes
• MetaboAnalyst — https://www.metaboanalyst.ca
• eggNOG — https://eggnogdb.org
• NIAGADS Open Access — https://www.niagads.org
• FlyWire — https://flywire.ai
• Ensembl — https://www.ensembl.org
• UCSC Genome Browser — https://genome.ucsc.edu
• ClinVar — https://www.ncbi.nlm.nih.gov/clinvar/
• dbSNP — https://www.ncbi.nlm.nih.gov/snp/
• gnomAD — https://gnomad.broadinstitute.org
• GWAS Catalog — https://www.ebi.ac.uk/gwas/
• Open Targets Platform — https://platform.opentargets.org
• GTEx — https://gtexportal.org
• ENCODE — https://www.encodeproject.org
• RegulomeDB — https://regulomedb.org
• JASPAR — https://jaspar.elixir.no
• ReMap 2022 — https://remap.univ-amu.fr
• cBioPortal — https://www.cbioportal.org
• EMDB — https://www.ebi.ac.uk/emdb/
• PRIDE Archive — https://www.ebi.ac.uk/pride/
• Human Protein Atlas — https://www.proteinatlas.org
• g:Profiler — https://biit.cs.ut.ee/gprofiler/
• Guide to PHARMACOLOGY — https://www.guidetopharmacology.org
• Mouse Phenome Database — https://phenome.jax.org
• Paleobiology Database — https://paleobiodb.org
• WoRMS — https://www.marinespecies.org
• NCBI BLAST — https://blast.ncbi.nlm.nih.gov

Notable conditions: UniProt and the RCSB PDB are CC-BY-4.0 and require attribution; AlphaFold DB is CC-BY-4.0 with EMBL-EBI's terms; ChEMBL is CC-BY-SA-3.0, which is share-alike; Europe PMC content is governed by each article's own licence, which varies per record; MetaboAnalyst is free for academic use, and its MIT-licensed R package does not cover commercial use of the hosted service. PubChem and NCBI resources are generally public domain but subject to usage rate policies.

Where a database's terms require a licence for commercial use, that requirement applies to you regardless of the fact that you reached it through Corollary.

nf-core pipelines are MIT-licensed. The tools they orchestrate are not: a Nextflow pipeline pulls dozens of third-party bioinformatics tools, some GPL, some academic-only, and a few requiring a separate commercial licence. The pipeline's licence is not the licence of what it runs.

The code sandbox installs open-source Python and R packages from PyPI, conda-forge, Bioconductor and CRAN. Those carry their own licences, including copyleft ones, and anything you build with them inherits their conditions.

The Corollary name, logo, interface and application code are ours or our licensors'. Nothing on this site grants you a right to use our trade marks. The interface is built on open-source components under permissive licences; their notices are preserved in the distributed bundle.

If you are the author or rights holder of anything listed here and the licence is recorded wrongly, or you would like an entry removed, write to legal@corollary-labs.com and we will correct or remove it promptly. We would rather be told than be right.