Visual biology model
A foundation model trained on 2 billion cell images. It recognizes more than 2,500 molecular activities, such as protein aggregation, pathway signaling and mRNA regulation, directly from imaging.
BioGPU Platform
Where AI agents run real biology
BioGPU pairs a visual biology foundation model with automated cell experiments, so AI agents can see how molecules change cellular networks and use that to design better ones.
2B
training images
1,000+
experiments per run
Millions
of cells across conditions
2,500+
molecular activities
Core components
Three layers, one runtime
Visual Biology Tokens
Every observation becomes a token that links the activity to its biological meaning and the experimental evidence behind it. Agents reason over these tokens the way language models reason over words.
BioGPU Agent
An AI agent that designs experiments, runs them on the platform and learns from the results. It covers 140 cell types and 25 diseases.
How it works
A closed learning loop
Every experiment advances the next. Agents go from question to evidence in cycles rather than months.
Ask
An agent or scientist poses a question, such as which compounds reduce TDP-43 accumulation without disrupting mTOR signaling.
Design
The BioGPU Agent selects cell types, conditions and molecules, and plans an experiment of 1,000 or more wells.
Observe
Automated imaging captures millions of cells. The visual model turns each image into Visual Biology Tokens.
Learn
Results update the agent's understanding and feed the next round of molecule design.
Mechanism-level insight
See the biology behind every hit
Activity scores alone don't explain a compound. BioGPU shows which pathways a molecule touches and which it leaves alone.
- Readouts across many molecular activities at once
- Early flags for off-target effects that can affect safety
- Every claim traces back to images and experiments
Pharma workflows
Built into how discovery teams already work
Target biology
Map how a target behaves across cell types and disease states.
Hit finding
Screen libraries against cellular mechanisms, not just binding.
Lead optimization
Guide chemistry with mechanism and safety signals.
Agent integration
Connect your own AI agents to BioGPU as a biology runtime.
See BioGPU on your biology
We'll walk your team through a workflow that fits your targets.