AI and quantum
for the human genome.

The lab. You get a persistent workspace with Python and notebooks, an agent that reads and writes the code with you, a versioned dataset store, free quantum simulators at 0 credits, and a sealed, reproducible record of every run. Point it at a locus and it gathers the public evidence, writes the analysis, runs it and shows its work. What is not live is the Substitution Atlas, a precomputed map of predicted variant effects across the genome — that is in development, and this page never pretends otherwise.

Not yet, and we will not say we do. Today the agent reasons over public records, published literature and open models, and runs the analysis you approve. A sequence-to-function model of our own is in development, and access to it opens with the Substitution Atlas. Everything on this page that is live is labelled live.

In small, hard subproblems inside a classical pipeline — a combinatorial solve, a kernel evaluation, a few-orbital electronic-structure calculation. The published record is proof-of-concept scale: toy assembly instances on annealers, quantum kernels reaching parity with classical models on reduced datasets, lattice-model folding of short chains. No result shows a quantum advantage for genomics at any realistic scale. We put a quantum step in front of you when it is the honest tool for that subproblem, we run it on the free simulator first, and we tell you what it cost against the classical baseline.

Only if you send it. On the desktop app the agent can profile a tabular file in place — schema, column ranges, missing-value counts, a SHA-256 of the bytes — and only that profile comes back. When you do upload, the dataset is versioned and sealed: re-uploading creates a new version rather than overwriting, the runner verifies the bytes against the sealed hash on arrival, and a run whose data does not match refuses to execute. Note that this is a data-custody design, not a regulatory certification: we make no HIPAA or GDPR compliance claim for your use of it.

Each run freezes a content-hashed manifest of its code, its declared dependencies, its inputs and the backend, so the same inputs always produce the same fingerprint. Paid runs come back with an execution receipt sealed by the executing service under a SHA-256 Merkle root. Either way this proves integrity and reproducibility, not origin, and we describe it that way.

Nothing. Quantum simulation runs at 0 credits on every plan and free compute runs are included each month. Paid hardware — real QPUs and GPU workers — asks for an estimate and your approval before a credit moves, then settles for what was actually used. You cannot be surprised by a bill you did not approve.

No. Nothing here is a diagnostic device and no output is medical advice. It is a research instrument: it helps you gather evidence, model a hypothesis and keep a reproducible record. Clinical interpretation stays with qualified people and validated assays.

Through the API platform and the Python SDKs: the same backends, the same balance and the same sealed records the chat uses, with scoped keys and a queryable ledger. Driving the lab from your own agent over the Model Context Protocol is built and opening with early partners rather than self-serve today — ask and we will tell you where it stands.