Your discoveries happen faster. And they stay yours.
A desktop application: the notebook your lab runs on, an agent that does the bioinformatics and logs the code it ran, and a model trained on measured biology. All of it on your machine.
Trusted by researchers at
Why this exists
Most of biology has never been written down.
Every public experiment ever run is still sitting there as measurements. What was published from them is the question each study went looking for — a thin slice. The relationships nobody thought to ask about are still in the data, unwritten and unsearched. That is the space we build models for.
A paper is a slice of an experiment
Studies report what they set out to test. The rest of the measurement stays in the data, never written up, and no search will find it.
A model of text stops where text stops
Language models are bounded by what has been published. Scale changes how well they read that record — not what is in it.
Your own files can only restate themselves
Your data holds what you measured and what you named. The connections between your datasets, and to everyone else’s, live outside them.
What is Heureka?
A biology model, a scientific agent, and the notebook they work in.
The model is the reason it works. The notebook is where it happens — a desktop application you install, not a website you log into, with your data on your own disk.
Archimedes
The biology model
A proprietary model trained on hundreds of thousands of real biological samples — not text — so Heureka can place, QC, and contextualize your data in ways a general LLM can’t.
ARC
The scientific agent
Ask in plain language and ARC does the work — bioinformatics, statistics, figures, literature — showing every step and logging it with the code that produced it.
Heureka Bench
The notebook
Where all of it happens: experiments, datasets, papers, protocols and inventory in one local-first workspace — and the reason the model can work on your actual files.
Why not just ChatGPT or Claude?
A general assistant reasons over text. Heureka reasons over your data.
A general AI assistant
Even science-tuned assistants are bounded by what has already been written down. They are excellent at summarizing the literature and drafting text — but they reason from published conclusions, not from measured biology.
Heureka
Adds what a chatbot structurally can’t: Archimedes, a model trained on real biological data, running inside a local-first notebook that works on your actual files. It places your samples against hundreds of thousands of real ones, flags mislabels, matches lab models to patient samples, and prioritizes the experiments worth running — every step logged and reproducible, on your machine.
All of biology, in one space
Archimedes learns the structure of human biology from hundreds of thousands of real experiments — not text — and places any sample into a single, batch-free space. Data from different platforms, labs, and studies becomes directly comparable — so you can stratify cohorts, connect molecular features across omics, and match lab models to patients in one place.
Data-driven
Learns from measured biology, not published text
Batch-free
One platform- and lab-invariant biological space
At scale
Grounded in hundreds of thousands of curated samples
Private & yours
Build a bespoke model on your data — your IP stays yours
What you can ask it — live in Heureka Bench
Stratify any sample
Place an expression profile in biological space and read its likely tissue, disease, and nearest cohort.
Flag mislabels and novel states
Catch when a sample doesn’t match its claimed label — quality control, or the first sign of something new.
Explore across omic layers
Genes, methylation sites, and microRNAs share one connected space — see what relates to what.
Connect molecules to phenotypes
Link a gene, methylation site, or microRNA to the tissues and diseases it marks.
Bridge patients to models
Route a tumor or patient sample to its most similar reference cell lines for follow-up experiments.
Grounded, with confidence
Every answer is tied to measured biology and its own confidence — what the model knows, not a guess.
Live in Heureka Bench
Ask in plain language inside ARC. Archimedes runs as a set of tools and returns structured, sourced answers — every result tagged with its confidence and caveats.
Your data, your model
For teams with proprietary data, we build a private, bespoke model on top of Archimedes. Your data stays isolated — never shared, never used to train anyone else’s model — while benefiting from our corpus and training infrastructure. A durable research asset that compounds as your data grows.
Research
We build our own models — and the data they learn from.
Heureka is a research company first. Archimedes is the model you can use today. It isn’t the only one we’re building.
Biological representation
One space where expression, methylation and microRNA become directly comparable — learned from measured experiments rather than from what has been written about them.
Molecules and their targets
Sequence-level models of how a molecule meets a protein, and where on that protein it lands — for pairs nobody has ever measured.
Perturbation and response
How a cell changes when you intervene, and how far that carries to samples you can never run the experiment on.
Some of this is in Bench today. More coming soon.
Your lab, in one place.
All your research, one place
- Organize experiments, datasets, and analyses by research question
- Local and cloud storage with workspaces
- Agentic search across all your data and projects
Curate your paper library
- Import by DOI — metadata and abstract auto-extracted
- Read and annotate PDFs inline
- AI summaries, methods extraction, and linking to your projects
See how your ideas connect
- Notes with bidirectional links
- Visual knowledge graph across your research
- AI digests surface patterns and connect ideas you might miss
Cloud compute, from your desktop
- Run heavy analyses without leaving Bench
- Select data and define the job
- Results sync back to your project
And everything else the lab runs on
Lab inventory
Reagents, consumables, equipment and animal colonies in one view, with automatic low-stock alerts.
Sequencing, ready to analyze
Pull FASTQs straight from FTP/SFTP and run scRNA-seq or metagenomics pipelines.
Scheduled tasks
Literature scans, hypothesis generation, and analyses that run on their own schedule.
Findings to slides
Build decks from your research — figures update when the underlying data does.
Published research
Work built with Heureka
“Targeting complex II leads to rapid disease regression and extends survival — a promising therapeutic target in AML.”
“AI helps us identify mechanisms of action and the patients most likely to respond. For biotech, it isn’t an advantage — it’s a necessity.”
Privacy & security
Your data, your machine
Local-first
Your files, data, and analyses stay on your disk. The cloud is opt-in, per job.
Privacy Mode
Lock a project and every data-egress tool is blocked at our server — only the model call leaves.
Never trained on your data
Your data stays isolated and private — including any custom model we build on it.
Download Heureka Bench
No subscription. Your data stays on your machine.