Corollary vs Biomni and Biomni Lab

The short answer

Biomni comes in two forms. The open-source agent from Stanford is a Python package you install on your own Linux machine and run with your own language-model key. Biomni Lab is the hosted product from Phylo, the company its creators founded: a web app with cloud compute, licensed data and paid plans.

Corollary is closest to Biomni Lab: a hosted agent with a machine of its own, public databases, bioinformatics programs and nf-core pipelines. The differences are in what comes with it and on what terms. Corollary runs 145 open scientific models with no GPU or key of your own, and it is free — but it is a non-commercial research demonstration, so it is not an option for company work.

Choose open-source Biomni to read and change the agent's code or to run it inside your own infrastructure, and Biomni Lab for a commercial contract, team workspaces or its licensed data. Choose Corollary to start in a browser today, free, with structure prediction, protein design and pipelines already set up.

Side by side

Biomni's column is from its own public pages, read on 24 October 2026; the numbers link to them below. Corollary's figures are counted from the platform itself and each one links to the page that lists them.

Where Biomni is the better choice

• You can read, change and self-host the open-source agent, and choose the language model it runs on.^[3][1]
• Biomni Lab is sold for commercial work, with enterprise deployments in a single tenant or in your own cloud. Corollary is a non-commercial demonstration.^[6]
• Biomni Lab takes uploads of up to 10 GB a file and imports from Google Drive, Dropbox and S3.^[9]
• Team workspaces with viewer, editor and admin roles.^[16]
• Licensed data that open tools cannot reach, such as COSMIC and Addgene.^[12]
• A peer-reviewed method: the agent is described in a paper in Science.^[5]

Where Corollary is the better choice

• Nothing to install and no API key: the agent, its machine and the language model come with a free account.
• 145 open scientific models — structure prediction, protein design, docking, genomics — run from the browser, with no GPU or key of your own.
• A published free allowance: 30,000 credits a month, and database queries never use credits.
• The machine is a full Linux box: the agent installs from bioconda, pip or a GitHub repository and runs whatever the work needs, your own scripts included.
• 143 nf-core pipelines, each with a page saying what it does, run as jobs on machines of their own.

Try the same work on Corollary

Ask in plain language. The agent works on a Linux machine of its own: it installs and runs what the job needs, keeps every file, and reads the results back with the numbers explained.

Questions

Sources

1. [1]Biomni — README (GitHub)
2. [2]Biomni — environment setup
3. [3]Biomni — licence (Apache-2.0)
4. [4]Biomni — commercial-mode data list
5. [5]Stanford HAI — Stanford scientists build an AI lab partner (9 July 2026)
6. [6]Phylo — pricing
7. [7]Biomni Lab docs — models and compute
8. [8]Phylo — how the Biomni agent uses high-performance computing (3 March 2026)
9. [9]Biomni Lab docs — FAQ
10. [10]Biomni Lab docs — pipelines
11. [11]Phylo — terms of service (6 July 2026)
12. [12]Phylo — launch announcement (3 February 2026)
13. [13]Phylo — Boltz partnership (16 June 2026)
14. [14]Biomni Lab docs — pricing FAQ
15. [15]Phylo — Biomni Lab
16. [16]Biomni Lab docs — collaboration