Send a genome.
Get back a signed report.
BioJarvis is the software layer between your sequencer and your clinician. It calls the variants, works out which ones matter, writes the report, and files it into your hospital system — in your country, on your keys. 250+ endpoints. One contract. One audit trail.
Scroll to descend · 30-day evaluation · No purchase order
Eighty-six billion neurons, and a question about one of them.
Connectome-scale modelling of neurodegeneration: Hodgkin–Huxley membrane kinetics, Wilson–Cowan population dynamics, and amyloid-β, tau and α-synuclein aggregation tracked over years rather than timepoints.
Cells are not the unit of disease. Neighbourhoods are.
Spatial transcriptomics, immune infiltration, stromal mechanics and the tumour microenvironment — where cell identity, position and neighbours have to be read as one signal rather than three.
Where genotype becomes behaviour.
Signalling integration across PI3K, MAPK, WNT and Notch; cell-cycle checkpoints, fate decisions and transformation risk. The virtual cell runs the perturbation and reports what the pathway does next.
A variant is only a hypothesis until the protein answers.
Structure, stability and splicing decide whether a substitution matters. AlphaFold confidence, ESM embeddings and deep-mutational evidence resolve the variants of uncertain significance that stall a report.
It begins as chemistry. Four bases, 3.4 ångström apart.
Germline and somatic calling against GRCh38 and T2T, benchmarked continuously on NIST Genome in a Bottle reference materials with the GA4GH comparison engine. Every clinical claim traces back to here.
One digital twin. One report a pathologist can defend.
Pharmacogenomics, polygenic risk, exposome and PK/PD converge into a single longitudinal model — and a signed, versioned clinical report that lands in the EHR your clinicians already use.
And then it is a public-health question.
Cohort analytics at biobank scale, pathogen phylodynamics, antimicrobial resistance and biosecurity surveillance — the scale at which a health ministry has to make a decision this week, not next year.
Same platform. Five different jobs.
Genomics touches every department and each one wants a different answer. Here is what BioJarvis means to each of them, without the jargon.
“My team is drowning in variants.”
A whole genome throws up five million variants. BioJarvis narrows that to the handful worth a scientist's time, with the evidence attached, so curation takes minutes instead of days.
“I want the result in my system.”
Reports arrive inside your EHR as structured FHIR — not a PDF in an inbox. Prescribing alerts fire at the point of order, before the drug is signed for.
“Where does the data actually live?”
Wherever you say. Managed, your national cloud, or fully offline. You hold the keys and the database, and every record access is logged and exportable.
“Prove it before we sign.”
Free 30-day key, no purchase order. Run our benchmark on your own reference material and get your own numbers. Open export formats and source escrow mean you are never locked in.
“We need national capability.”
Rather than funding a five-year build, deploy a working stack into your jurisdiction and staff it with your own people. We train them and hand over the keys.
“Can it answer research questions?”
Above the clinical layer sits simulation and causal inference — cohort analytics, digital twins and intervention modelling, on the same governed data.
It begins as chemistry.
Three billion base pairs, four letters, one reference. Before anything can be called clinical it has to be called correctly — aligned, genotyped, phased, and scored against a truth set somebody else produced. We start where the signal starts.
Every chromosome. Every reportable locus.
ACMG secondary findings, pharmacogenes, the hereditary cancer panel, the cardiomyopathy panel and the newborn screening set — drawn here at their true megabase coordinates on GRCh38, because a panel you cannot point to on a chromosome is a claim, not a capability.
One platform from base pair to biobank.
A variant matters because of a protein, which matters because of a cell, which matters because of a tissue, which matters because of a person — and sometimes because of a hundred thousand of them. Most tools pick one level and stop. We carry the evidence across all of them, on one governed copy of the data.
A variant never acts alone.
Pathogenicity is a network property. We pull protein interaction, pathway membership, expression context and regulatory architecture into the same call, so a novel missense is judged by the company it keeps rather than by a single score in isolation.
And where the genome becomes behaviour.
Electrophysiology, connectomics and neurodegeneration modelling sit on the same stack — because for a great many of the conditions our customers care about, the phenotype is measured in spikes and not in base pairs.
Everything between the sequencer and the clinic.
Most vendors sell you one slice — a caller, or an interpretation tool, or a reporting front end — and you spend two years wiring them together and losing the audit trail at every join. BioJarvis is all of it behind one login, one contract and one log. Take the whole thing or licence only the parts you need.
250+
documented endpoints, all under one API key
20
clinical and research domains
62
data planes behind one authentication boundary
3
deployment modes on the same codebase
It plugs into what you already run
No rip-and-replace. BioJarvis speaks the formats your instruments emit and the standards your hospital systems already accept.
Four things almost nobody else does
Not adjectives — things you can check before you sign.
We publish our gaps
Our validation page lists which capabilities are benchmarked against public truth sets and which are not, by name. Ask any other vendor for that table.
See #/validationYou generate the numbers
The benchmark harness ships with the free key. You run it on your own reference material, on your own hardware, and the figures you take to procurement are yours, not ours.
GA4GH hap.pySovereignty is a flag
Managed, national cloud or air-gapped run from the same codebase. Data residency is a deployment setting, not a twelve-month professional-services project.
3 deployment modesLeaving is designed in
Open export formats at any time, free of charge, and source escrow on national licences. We keep your business by being worth renewing, not by holding the data.
VCF · FHIR · PhenopacketsDiagnostic pipeline
Interoperability & governance
Simulation & causal reasoning
Decision support above the clinical layer. These modules are not diagnostic devices and are labelled as such in every response.
We publish what is measured, and what is not.
Procurement deserves a straight answer. Germline calling is benchmarked continuously against NIST Genome in a Bottle reference materials using the GA4GH hap.py comparison engine — the method a regulator would apply. Where a capability has no external truth set yet, this page says so rather than quoting a figure from the literature.
| Capability | Reference material | Method | Status |
|---|---|---|---|
| Germline SNV | GIAB HG001–HG004 | hap.py · GA4GH | Benchmarked |
| Germline indel | GIAB HG001–HG004 | hap.py · GA4GH | Benchmarked |
| Confident-region recall | GIAB high-confidence BED | stratified | Benchmarked |
| Somatic SNV | SEQC2 truth sets | planned | In validation |
| Structural variants | GIAB HG002 SV benchmark | planned | In validation |
| PGx star alleles | CDC GeT-RM | planned | In validation |
| Simulation modules | — | n/a | Decision support |
How the benchmark runs
Reference material
NIST GIAB truth VCFs and high-confidence BED regions for HG001–HG004 are pulled from the NCBI FTP and pinned by release version.
Pipeline run
Your pipeline configuration processes the sample end to end — alignment, calling, filtering — with every tool version recorded.
Comparison
GA4GH hap.py compares the call set against truth inside the confident regions, stratified by variant type.
Report
Precision, recall and F1 for SNV and indel, returned as a versioned artefact you can attach to a procurement submission.
Benchmarks run on demand against your own reference material during evaluation, so the figures you procure against are produced on your data and your hardware — not quoted from ours. We do not publish a headline accuracy number, because a number produced on our hardware with our configuration would not predict yours.
Your genomes stay where your law says they stay.
Three deployment modes on one codebase. Sovereignty is a configuration, not a bespoke engagement — the same build runs managed, in your national cloud, or fully air-gapped.
Managed cloud
Operated by Bright Group in your chosen region. The fastest route to production for departments without platform engineering capacity. Patching, backups and monitoring included.
Sovereign tenancy
Deployed into your national cloud or private datacentre. You hold the encryption keys and the database; we hold no copy of patient data at any point in the lifecycle.
Air-gapped
Containerised install with offline reference bundles for defence, biosecurity and classified programmes. Knowledge bases update by signed physical media on your schedule.
Built for the inspection
Isolation by default
PostgreSQL row-level security per tenant, TOTP multi-factor authentication, scoped role-based access and an append-only audit log covering every record touched.
Laboratory quality
CLIA QC lots with Levey–Jennings review, CAPA workflow, proficiency-testing records and per-run quality gates with defined acceptance thresholds.
Consent & ethics
Tiered consent with withdrawal, IRB protocol and enrolment tracking, and a governance engine that evaluates intervention designs against policy before they execute.
Data-subject rights
Export, rectification, restriction and erasure workflows for GDPR and equivalent regimes, with report versioning so an amendment never overwrites a signed original.
Provenance
Every intermediate observation traces to the source, the reference build and the tool version that produced it — so a reclassification three years later has a defensible lineage.
Standards
GA4GH VRS and Phenopackets, HL7 FHIR Genomics and v2 ORU, SMART on FHIR and CDS Hooks — the interfaces your inspectors and your EHR vendor already recognise.
On certification. BioJarvis provides the workflows and evidence trails that CLIA, CAP, GDPR and equivalent regimes require. It does not itself hold those accreditations — accreditation attaches to your laboratory and your deployment, and we support that submission rather than substitute for it.
Evaluate free. Licence annually. Own it outright.
Evaluation is genuinely free and needs no purchase order — a sandbox key, the full endpoint surface and GIAB reference data, so your bioinformatics team can benchmark the platform before anyone signs anything.
Evaluation
Single user · 30 days- ✓Sandbox key, every endpoint
- ✓50 samples, synthetic and GIAB data
- ✓Clinician and patient dashboards
- ✓Run the benchmark suite yourself
- ✓No PHI, no purchase order
Department
Single site- ✓Production pipeline with sign-off
- ✓Variant, PGx, PRS, rare disease
- ✓FHIR and HL7 v2 into your LIS
- ✓Managed cloud in your region
- ✓Business-hours support, named engineer
Institution
Multi-site · accredited- ✓Everything in Department
- ✓Somatic oncology suite
- ✓Single sign-on, tenant isolation
- ✓CLIA/CAP QC, CAPA, IRB workflows
- ✓Sovereign tenancy, contracted SLA
National
Sovereign programme- ✓Biobank-scale ingestion and cohorts
- ✓Air-gapped or national-cloud deploy
- ✓Biosecurity and pathogen surveillance
- ✓Custom data planes, co-development
- ✓Training, secondment, source escrow
What each tier includes
| Capability | Evaluation | Department | Institution | National |
|---|---|---|---|---|
| Endpoint access | All | All | All | All |
| Sample volume | 50 | Contracted | Contracted | Biobank scale |
| Patient data (PHI) | — | Yes | Yes | Yes |
| Clinical sign-off workflow | Demo | Yes | Yes | Yes |
| Somatic oncology | Demo | — | Yes | Yes |
| Single sign-on | — | — | Yes | Yes |
| Sovereign tenancy | — | — | Optional | Yes |
| Air-gapped deployment | — | — | — | Yes |
| Biosecurity module | Demo | — | Optional | Yes |
| Source escrow | — | — | — | Yes |
| Support | Community | Business hours | Contracted SLA | Dedicated team |
Procurement questions
Why is pricing on application rather than published?
Licence cost depends on sample volume, deployment mode, module scope and support level, and these vary by an order of magnitude between a single department and a national programme. We quote against your actual scope rather than publish a figure that would be wrong for most readers. Evaluation is free and needs no quotation.
Is BioJarvis a regulated medical device?
No. BioJarvis is decision-support software. Clinical outputs require review and sign-off by a qualified professional, and the simulation and inference modules are explicitly labelled as decision support in every response. Where a deployment forms part of a regulated diagnostic service, that accreditation attaches to your laboratory; we support the submission.
What happens to our data during evaluation?
Evaluation uses synthetic and public GIAB reference data only — no patient-identifiable information should be uploaded, and the sandbox is provisioned without a PHI pathway. Production tenancies isolate data per tenant at the database row level.
Can we run our own benchmark before committing?
Yes, and we would rather you did. The evaluation key includes the benchmark harness. Point it at your own reference material and it runs GA4GH hap.py against your call set on your hardware, returning precision, recall and F1 as a versioned artefact you can attach to a submission.
What if we need a module that does not exist yet?
National-tier engagements include custom data planes and co-development. The platform is organised as sixty-two data planes behind one authentication boundary, so a new domain is usually an ingestion and schema exercise rather than a new product.
What are the exit terms?
Your data is exportable in open formats — VCF, FHIR bundles, Phenopackets — at any time and without charge. National-tier licences include source escrow. We do not hold data hostage as a renewal mechanism.
Benchmark us before you brief procurement.
Tell us the programme and we will provision a sandbox key, load GIAB reference material, and walk your bioinformatics lead through running the validation suite against it. Typical turnaround is two working days.
You submit
Institution, scope and data-residency requirement. No purchase order, no card, no call required.
We provision
Sandbox key scoped to your organisation, both dashboards, and GIAB reference material preloaded.
Walkthrough
Forty-five minutes with an engineer, screen-shared, running the benchmark on your material.
You evaluate
Full endpoint surface, your own tests, your own numbers. We answer questions; we do not chase.
Live environments
Open now, no key required — the API reference is fully browsable and both applications run in demonstration mode.
Fourteen clinical and research areas, one platform.
Pick the problem you actually have. Each area lists in plain language what it does, who uses it, the live endpoints behind it and the public reference data it draws on — so you can check the claim before you talk to us.
What it actually does
Who uses it
Live endpoints
Every one of these is in the public OpenAPI schema and callable with an evaluation key today.
Related areas
Biology is a pattern that computes itself.
The field behind this page is a Gray–Scott reaction–diffusion system running live on your GPU. Two chemicals diffuse at different rates; one feeds, one decays. From that alone, structure emerges — spots, stripes, labyrinths — with no template telling it what to become. Move your cursor and you seed it.
Turing, 1952.
Alan Turing proposed exactly this system as the chemical basis of morphogenesis — how an animal gets its markings and a body plan gets its axes. It is the oldest formal answer to the question BioJarvis exists to answer at scale: how does a genome become an organism, and where in that process does a variant actually matter?
∂u/∂t = Du∇²u − uv² + F(1−u)
∂v/∂t = Dv∇²v + uv² − (F+k)v
F = 0.037 · k = 0.060 · Du = 0.21 · Dv = 0.105 — the coral regime
Structure without a blueprint
No gene encodes a leopard's spots directly. It encodes a reaction, and the pattern falls out of the physics. Interpretation has to model the process, not just read the code.
Small parameters, large consequences
Shift the feed rate by a few thousandths and spots become stripes become chaos. That sensitivity is why a single base substitution can end a pregnancy or do nothing at all.
Scale is not a metaphor
The same mathematics describes pigment cells, cortical folding and epidemic waves. A platform that reasons at one scale only is not reasoning about biology.
Watch a genome actually get analysed.
A demonstration run on NIST GIAB HG002, replayed from a completed job — alignment through calling, annotation, ACMG classification and benchmark. Every stage names the tool and the version that produced it, which is what a provenance chain looks like in practice.
Demonstration sequence · not a live production job
We got tired of watching the same build fail.
Every national genomics programme starts the same way: a caller from one vendor, an interpretation tool from another, a reporting front end from a third, and two years of integration work in between. Then a variant gets reclassified, someone asks which run produced the original call, and nobody can answer. We built the whole path once so that question always has an answer.
Measured, not asserted
We publish a validation table that says which capabilities are benchmarked against external truth sets and which are not. Most vendors publish an accuracy number and no method.
Sovereign by construction
The same codebase runs managed, in your national cloud, or air-gapped. Data residency is a deployment flag, not a twelve-month professional-services engagement.
Standards first
GA4GH, HL7 and NIST are not integrations we added later — they are the interfaces the system was designed around, because that is what makes it inspectable.
The standards your inspectors recognise
BioJarvis reads from and writes to the public reference infrastructure of clinical genomics. These are the bodies whose formats and reference materials the platform depends on.
Talk to an engineer, not a sales team.
Evaluation requests are read by the people who built the platform. If your question is technical, you will get a technical answer.