Questions, answered straight.

NOVA-3 is a computer-aided design suite for multi-specific antibodies. It assembles candidate molecules from a curated, schema-typed Module Library of validated parts, ranks them in silico, refines the hard chemistry with quantum active-space corrections, and ships every result with a cryptographically anchored Evidence Bundle. The thesis behind it: the molecules worth making are designed, not screened.

A folding model predicts the 3-D structure of a sequence you already have. NOVA-3 starts a step earlier — it composes a multi-arm molecule from validated modules to hit a target specification, then uses folding models (which it wraps) to evaluate the candidates. We design the molecule; folding models predict a structure for it.

A multi-specific binds two or more targets at once — for example a trispecific T-cell engager, a conditional costimulation gate, or a blood-brain-barrier shuttle. Single-target antibodies are commoditizing; the molecules pharma now pays for are combinatorial, and they are exactly the ones a parts-based design approach helps with. See the modality gallery.

No. We don't out-fold them and don't claim to — we wrap best-of-breed open foundations and add the layers they don't ship: multi-specific composition, cryptographic provenance, quantum-corrected refinement, and a wet-lab data flywheel. The honest comparison is on the Benchmarks page.

Yes. The Predict console and Playground call ESMFold live and render the returned structure in the Mol* viewer in your browser, coloured by pLDDT confidence. Nothing is pre-rendered — paste a different sequence and you get a different fold.

No login, no signup. The demo runs in your browser against a public folding endpoint. Most tools in this space gate the viewer behind a login — we deliberately don't.

Folding is a real model inference over a public endpoint, so latency depends on sequence length and load, and very long sequences can time out. If a fold fails, try again or shorten the sequence. This is the genuine compute cost of structure prediction, not a loading animation.

They are pLDDT — per-residue model confidence, on the AlphaFold convention: blue is very high (≥90), light blue confident (70–90), yellow low (50–70), orange very low (<50). The legend under the viewer names each band. High confidence means the model is sure of that region's local structure; it is not a measure of binding or function.

Three internal programs at different design stages: P1 — a CLDN18.2 × CD3 × 4-1BB trispecific T-cell engager (in-silico PoC); P2 — a TfR1 / IGF-1R blood-brain-barrier shuttle (in-vitro PoC); P3 — a GD3 × IL-7R senolytic bispecific, our flagship (lead candidate ranked). Details on the Pipeline page.

Not yet — and we're explicit about it. The programs are at design and in-silico stages; wet-lab validation (SPR/BLI affinity, ADCC, and the P2 in-vivo study) is scheduled, not complete. The program stage chart shows exactly how far each has reached, with planned work marked as such.

No. The figures on the site are predicted (in-silico) values and design targets, labelled accordingly. We have not represented any wet-lab measurement we don't have. Measured results will publish on Benchmarks as programs mature.

They are provisional patent applications (PROV-2026-001/002/003), not granted patents. A provisional establishes a priority date and gives twelve months to file a full application; it is not an issued patent, and we don't describe it as one.

A reproducible record attached to every result. The inputs, commands and outputs are canonicalized, each artifact is SHA-256 hashed, the hashes are combined into a Merkle root, and that root is timestamped and anchored to a public chain. The provenance flowchart walks through it step by step.

We use OpenTimestamps to commit the Merkle root to the Bitcoin chain. A public chain means there is no trusted timestamper to take our word for — anyone can independently verify that a given root (and therefore the design behind it) existed at or before a specific block. That gives pharma diligence teams a notary-grade, tamper-evident invention date.

Ownership of designs created for a partner's target is set in the engagement agreement, and is typically assigned to the partner; QAI Labs retains its underlying platform, Module Library and methods IP. Specifics are documented per engagement — we're happy to walk through the framework under NDA.

No. We make no claim of demonstrated quantum advantage in ground-state chemistry — an open research question we engage with honestly rather than market around. The restraint is deliberate, and we'd rather under-claim than over-claim here.

Targeted active-space corrections on hard chemistry — for instance the glycan binding pose in P3 — via CVaR-VQE, plus QAOA for some assignment problems. It is a small, corrective slice of the compute stack, not the workhorse; classical GPU carries the bulk. The compute allocation chart shows the split.

Jobs fall back to a simulator, and every job logs its backend, shot count, calibration snapshot and cost so the hardware fraction is always transparent. Nothing in the core workflow depends on a specific machine being online.

Access is role-based and least-privilege. Partner programs are segregated with IP carveouts so one engagement's data and modules aren't visible to another. Full detail is on the Security & Trust page.

Module-level IP carveouts and access controls keep each partner's inputs, derived modules and results walled off. The Module Library distinguishes shared, public-exemplar entries from partner-restricted ones at the schema level.

Encrypted at rest and in transit (AES-256 / TLS 1.2+), behind access-controlled infrastructure. See Security & Trust for the current controls and our Privacy Policy for data handling.

Submit a target through the form on the homepage or email partner@qailabs.co. The engagement flow shows what happens next — intro and mutual NDA, target scoping, a design scope back to you, ranked candidates, wet-lab validation, and delivery.

Engagements are scoped to the program — there are tiers from platform access through full design-and-validate partnerships. We quote against your target and goals rather than list a one-size price; see Pricing for the model and reach out for a scoped quote.

Yes. We can route a mutual NDA before you submit anything sensitive — just select that option on the partner form, or say so by email. We'd rather you share specifics under NDA than hold back.

QAI Labs Ltd. is an early-stage company — a Delaware corporation headquartered in Sugar Land, Texas. The Company page lays out the thesis, what's built today, and the roadmap, with future work clearly marked as planned.

Partnerships and general inquiries: partner@qailabs.co. Press: press@qailabs.co. Investment and corporate inquiries are welcome at partner@qailabs.co — more on the Contact page.