Lab-in-the-loop
Built for the future of AI in the lab
Something genuinely new is happening in the life sciences: software is starting to reason about experiments. We think the instruments on the bench should be ready for that — so Spatial Station and Cell Forge both speak ML-Prep API, an open REST interface over Ethernet that lets your own software design, refine and run protocols right alongside your team.
Working together
Your science, your software, one bench
-
Design
Protocols are structured data — panels, incubations, washes and timings. Anything that can shape that data can compose a protocol, so your software designs alongside you rather than waiting for you to type it in.
-
Refine
Adjust an incubation, swap a reagent, reshape a panel. Iteration happens in the place your thinking happens, and the instrument keeps up.
-
Run
Send it to the bench over the network, and let the instrument do what it does best — the same careful, repeatable work, every single time.
Closing the gap between insight and experiment
Today a spatial or flow experiment runs in a straight line: design a panel, run the samples, study the results — and then someone carries what they learned back to the bench and sets it up again by hand. The thinking is quick. It's the journey back to the bench that takes the weeks.
When the instrument is part of the conversation, that journey gets much shorter. A titration can come back and go straight out again at corrected concentrations. A screen can narrow itself between rounds. An analysis that finds a marker uninformative can hand back a better panel, ready to mix.
That's the loop we mean — discovery moving at the speed of the question, rather than the speed of the paperwork between them.
Ready for the agents you're already building
ML-Prep is an ordinary REST API, so anything that can make an HTTP request can work with it — including an AI agent. That turns out to be a natural fit rather than a clever trick: protocols are structured data, and structured data is exactly what today's models are good at reasoning about and improving.
We've built a Claude skill for designing and editing Parhelia protocols, and we'd love to show you what it can do. Come and talk to us.
Trustworthy by design
Working this way shouldn't mean giving anything up. An agent takes exactly the same path through the instrument that your operators do — no shortcuts, no side doors.
-
Authenticated
Every request arrives on a password-protected account with its own credentials. There is no unauthenticated route to an instrument, and never was.
-
Validated
The instrument checks every protocol before it will run one. The API cannot ask for anything the instrument wouldn't accept from its own touchscreen.
-
Recorded
Every run and every result is written to the instrument's internal database, alongside the run logs and user accounts your team already relies on.
So an agent-initiated run is every bit as accountable as one started at the touchscreen: same credentials, same checks, same record. That's what makes this something you can bring into a facility that has to show how every sample was prepared.
Where we fit
- What we bring
- Two shipping instruments that any program — or agent — can work with to design, refine and run protocols, with the reliability of a platform already at the bench in labs around the world.
- What you bring
- The science. Your analysis, your models and your judgement about what deserves to run next — because the interesting decisions in a laboratory should still be yours.
Let's build this together
If you're connecting analysis to a bench — or just thinking about what an AI-native lab could look like — we'd genuinely love to hear from you. These are the conversations we enjoy most.
Start a conversation