A visual platform that lets regulated labs orchestrate many AI models into governed, auditable workflows - without the data ever leaving the building.
Somewhere in a cancer lab, a scientist who has never written a line of Python is chaining together nine of the most advanced AI models on earth. She does it by dragging boxes across a screen. One box is AlphaFold2, predicting how a protein will fold. Another scores candidate binders. A third visualizes the result so she can eyeball the quality herself. The whole pipeline runs inside her institute's own firewall, and every step leaves a receipt. This is Salt AI on an ordinary Tuesday.
Most AI companies are racing to build a single, smarter model. Salt AI took the opposite bet: the models we already have are extraordinary, but they are scattered, hard to combine, and impossible to trust in a regulated setting. Salt's job is not to be the smartest voice in the room. It is to be the conductor that gets all the voices to play in time.
The company calls its category contextual AI for regulated enterprise. In plain terms: your models, your data, your infrastructure, your audit trail. For pharma and healthcare - industries where "where did the data go?" is the first question, not the last - that combination is the whole ballgame.
No single AI model can unlock the future of medicine alone. The future belongs to ensembles - models working in synergy.
It is the difference between an experiment you run this quarter and one you shelve. Salt re-engineered AlphaFold2 with a split-compute design that balances CPU and GPU, a 3TB cached dataset, and a GPU-accelerated search that turns multi-hour lookups into seconds - all while keeping full prediction accuracy.
A sample of the swappable, versioned research models in Salt's library.
Models, data, governance, and context in one deployable platform - model-agnostic, cloud-agnostic, vendor-agnostic. Runs behind the firewall with no public data egress and immutable logs for every prompt, weight, and response.
A visual-first IDE that lets non-coders - biochemists, oncologists, clinical teams - build and swap AI models into workflows and collaborate on them in real time.
A curated, versioned catalog of life-sciences models you can swap in and out mid-workflow - from protein folding to docking to diffusion.
Connects clinical and bioinformatics data into workflows while respecting data residency and sovereignty across jurisdictions.
Since the summer of 2024, the Ellison Medical Institute - whose drug-discovery work is led by physician-author Dr. David Agus - has used Salt to design and analyze thousands of compounds. As of August 2025, two protein candidates had advanced to in-depth wet-lab studies after positive in-vitro results. The live hit-identification pipeline chained nine models, paired with interactive quality-control visualizations so biochemists could inspect each fold. Implementation that once took months collapsed into weeks.
Salt's platform is also in use at UCLA's Bioengineering Department and the healthcare organization Hiteks, with the company signaling expansion into financial services, energy, and government - anywhere the AI has to be powerful and provable at the same time.
Salt was founded by CEO Aber Whitcomb, a co-founder and CTO of Myspace, and CTO Jim Benedetto. The founders describe a twenty-year track record of building high-performance computing and AI together. In November 2025, the company stacked its leadership bench with biotech and AI veterans to push into life sciences.
Nate Beyor previously led Health Tech at Boston Consulting Group - 20+ years across biotech, medtech, and healthcare.
Announced September 2025, led by Morpheus Ventures.
Across three rounds since 2023.
Struck Capital, Marbruck Investments, CoreWeave, Bill Tai, Brian Venturo, Irregular Expressions.
The deepest idea inside Salt is not the 22x speedup - it is the receipt. A model that recommends a film can be a mystery. A model that helps design a drug cannot. Regulators, and the scientists who answer to them, need to trace exactly how a prediction was made, validate each step, and prove the data never wandered off. Salt is built so that visibility is the default, not a feature you bolt on later.
That is why the platform runs on-premises or in a private cloud, why it logs everything immutably, and why it arrives pre-validated against SOC 2 and healthcare controls. It is unglamorous plumbing in service of a glamorous goal: getting good science to patients faster, without asking anyone to trust a machine they cannot inspect.
Salt is built to provide complete visibility into every step of the discovery pipeline.
Product demos, interviews, and the blog where Salt shows its work.
Back in that cancer lab, the biochemist finishes her pipeline. Nine models ran. Two candidates are already in the wet lab. She never left her institute's firewall, never wrote a line of code, and can show anyone - a colleague, an auditor, a regulator - exactly how each prediction was made. A year ago this took months and a team of engineers. Now it takes an afternoon and a few dragged boxes. Salt AI did not invent the models. It just got them to play together, quietly, in time - and handed the baton to the person who actually knows what she's looking for.