All the science of a lab.
None of the walls.
Heureka is building the infrastructure for autonomous science.
With models trained on measured biology, a system of record for every step, and a network of partner labs, our autonomous labs form hypotheses, design and run experiments, and learn from every result.
Discovery that moves faster, scales further, and stays yours.
Trusted by researchers at
Biology is one of the hardest problems we have. We’re building the intelligence to take it on.
A single cell runs on tens of thousands of interacting parts, and most of what goes wrong in disease is still unexplained. Most drugs that enter clinical trials still fail, often for reasons that were measurable years earlier. Closing that gap takes more than faster software. It takes an intelligence layer across the whole of research: models trained on measured biology, a record of every result, and autonomous labs that reason over all of it, test what matters, and learn from every outcome.
This is the beginning of a new era for drug discovery and for medicine. We intend to help build it.
Sources: NIH National Center for Advancing Translational Sciences; Nguengang Wakap et al., European Journal of Human Genetics (2020).
The scientific method, running continuously.
Hypothesis, design, experiment, learning, and around again. An autonomous lab runs the whole loop itself, every hour of the day. Experiments go out to a network of partner labs that already run them every day, and the data comes straight back. Or run them at your own bench.
Hypothesize
Reads the data, the literature and its own earlier results, then commits to what it thinks is going on, ranked by how much each idea would change the question.
Autonomous lab · 24/7Design
Picks the one experiment that would settle it: the assay, the controls, how many samples, and what each possible outcome would mean.
Autonomous lab · 24/7Run the experiment
Matched to a partner lab that runs that assay every day, run, and the data sent straight back into the lab. Prefer your own bench? Run it there.
Partner labs · or your benchLearn
The results come back. It checks them against what it predicted, keeps what held up, strikes what didn’t, and writes it all down.
Autonomous lab · then again↺ every result feeds the next hypothesis
Rigor, built into the machine.
An autonomous lab works the way a good lab does: researchers who each bring their own approach, a reviewer who checks the work, a PI who decides, and a notebook that misses nothing. Add a model of measured biology, and keep it running day and night.
In
Work the question
Review
Decide
Test
10+
One autonomous lab works like a team of ten or more researchers at full stretch.
Around the clock, every day, with no limit on how hard it works. It doesn’t replace your people. It gives each of them a team.
The record
A lab, not a single mind
Every question gets several researchers, each working it their own way. A finding counts when it holds up across them, and when it doesn’t, the lab says why.
Checked before it counts
Nothing reaches you until it has been reviewed against the data. What doesn’t hold up is struck, and the strike stays on the record.
Measured biology, built in
Archimedes, our model trained on real experimental measurements, places your samples among hundreds of thousands of them, so the lab reasons over data rather than over what has been written about it.
Experiments designed to decide
When the data can’t settle a question, the lab designs the experiment that can, with controls and a plan for reading the result before it comes back.
A notebook that misses nothing
Every step is written down as it happens and linked to what it builds on. Dead ends stay as visible as wins, and changing its mind is a new entry, never an edit.
Real code, in your project
The scripts behind every result, with the data, tables and figures they produced, live in your project. Anyone can open them and run them again.
Fail in a dish, not in a trial.
About nine in ten drugs that enter clinical trials never reach patients. Each approval carries an estimated $2.6 billion, most of it spent on the programs that failed. And the warning was usually there years earlier: an effect that held in one cell line and not the next, a mouse result that never carried over to people.
Heureka is built to find those failures while they still cost a plate, not a program.
first
Test the weak link first
It goes after the assumption most likely to break, while breaking it is cheap.
then
Ask whether it travels
Before a result moves up from cell line to animal, it is checked against measured biology across many contexts. What holds in only one gets flagged.
so
Spend on the survivors
Fewer, better experiments early, so the expensive stages only see what has already earned its place.
Sources: Sun et al., Acta Pharmaceutica Sinica B (2022), on the ~90% clinical failure rate; DiMasi, Grabowski & Hansen, Journal of Health Economics (2016), on the cost per approved drug.
$2.6B estimated cost per approved drug, most of it spent on the ones that failed.
Being right earlier is the cheapest way to change both numbers.
13 experiments nobody had to run.
The expensive part of research is picking what to test. We gave the same AI scientist the same 31 cancer cell lines, with and without Archimedes, and a budget of 30 experiments each. With Archimedes it reached the same result after 17.
experiments to the same result · 31 unseen cancer cell lines
AI scientist alone
AI scientist + Archimedes
One square per experiment. Green: what it took with Archimedes. Dashed: the 13 that never had to happen.
One autonomous lab. Many ways to use it.
A biotech’s first program, a core facility that never closes, a grant that runs every aim at once, a department that grows its research without growing its walls. Start a lab for the work you have, and add more as it grows.
For founders
Start your own biotech in five minutes.
Bring a question and a thesis. The lab does the work of an early discovery team: it pulls the data, builds the case, designs the experiments and sends them to partner labs. Your first program, before your first hire.
For biotechs
Spend on the experiments that matter.
Every experiment you don’t need is money and months back. In our benchmark, the lab reached the same result with 13 fewer experiments out of 30.
For every lab
Your bioinformatics core, always on.
Drop the data into the lab and get the analysis back: quality control, statistics and figures, with the code that produced them. For the lab without a bioinformatician, and the one whose bioinformatician has a queue.
For PIs
Your CRO, without the wait.
Hand off a project the way you would to a contract lab, and get a running program instead of a quote. One lab per grant aim, all moving at once, every step on the record.
For universities
Grow the research, not the building.
Most departments have more good questions than labs to pursue them. Give each group an autonomous lab working beside it, and the research a department does can grow without new space, equipment or hires.
Scale your science.
Every autonomous lab behaves like a lab: its own question, its own record, its own line of experiments. Start one for every question you never had the people for, and let them all run side by side, around the clock.
It does the science. You set the direction.
An autonomous lab doesn’t wait for instructions. It works its question day and night, sends the experiments it needs to partner labs and brings the data back. You approve each experiment with one tap. Everything else, it handles.
the brief
One short read, at your hour
What changed and why it matters. Every evening, once a week, or only when it needs you.
experiments
One tap, and it’s running
The design arrives ready to run. Approve it and the lab takes it from there: a partner lab runs it and the data comes back into the project. Or run it yourself.
steering
Redirect it whenever you like
Point it somewhere new, hand it a paper or a dataset, or ask what it has learned since Tuesday.
control
Pause it. Resume it. Stop it.
One click. Give it a stop condition and it pauses itself when it gets there. Its record stays yours either way.