BoltzFold
Queue protein folding predictions, monitor progress, and download structures.
commoditize protein folding
One API across commercial‑use protein folding models. Boltz‑2 (protein–ligand), OpenFold‑3 (AlphaFold3 analog, protein-only), and BoltzGen (binder design) are live now, and we keep adding modern backends as soon as they clear our quality and licensing bar. EU‑hosted, GDPR‑by‑default, reproducible.
Join the API waitlist
Tell us how you plan to use the BoltzFold API and we'll reach out when seats open.
Used by early teams at
How it works
1) Submit
POST /v1/predict with a FASTA (add ligand SMILES for Boltz‑2 protein–ligand). Or use the web UI.
2) We route
Auto‑mode selects the best model for speed vs accuracy (or pin one). Jobs run asynchronously on a serverless GPU backend.
3) Download
Get .pdb/.cif, confidence (pLDDT/PAE), and affinity (Boltz‑2). Every job includes a reproducibility manifest (model/weights hash, MSA DBs, seeds, GPU type).
Why teams switch to us
Multi-model, one endpoint
Boltz-2, OpenFold-3, and BoltzGen are available now. Route automatically or take full control — and expect new state-of-the-art backends as we validate them.
Predictable costs
Per‑residue pricing with one global price and prepaid credits; cache re‑use and budget guardrails. No tiers or enterprise plans.
Private by design
EU data residency, GDPR by default; zero‑retention mode, audit logs, signed manifests.
Developer-first
Simple REST, Python/JS SDKs, webhooks, standardized outputs; serverless GPU backend.
What's under the hood
- Models: Boltz-2, OpenFold-3, and BoltzGen — with more modern systems rolling out as soon as they meet our standards.
- API: Async jobs, webhooks, standardized artifacts (PDB/mmCIF, pLDDT/PAE, affinity), advanced binder design with BoltzGen
- Reproducibility: model+weights hash, MSA DB versions, seeds, GPU type
- Backend: Serverless GPU autoscaling across NVIDIA GPUs and CPUs
- Privacy: EU regions, GDPR by default, zero‑retention; no enterprise VPC/on‑prem (by design)
FAQ
Why do proteins matter?
Proteins are the building blocks of life. Their 3D shape determines what they do — how they bind, catalyze reactions, and signal. Knowing structure explains interactions and guides the design of drugs, enzymes, and antibodies. Traditional methods (X‑ray, cryo‑EM, NMR) are accurate but slow and expensive; fast in‑silico predictions help teams triage targets and iterate designs much earlier.
What are protein folding models?
Deep‑learning systems that predict a protein’s 3D structure from its amino‑acid sequence (and sometimes binding partners). Examples include AlphaFold, ESMFold, OpenFold-3, and Boltz-2 (protein–ligand). They output atom coordinates plus confidence metrics (pLDDT/PAE) — and in some cases, affinity estimates. Why this is transformative: minutes instead of months, lower cost, and reproducible outputs that plug into pipelines. They don’t replace experiments, but they dramatically accelerate them.
How fast are results?
Speed depends on sequence length, queue depth, and model choice. We share fresh benchmarks during onboarding and can tune sampling presets to fit your latency versus accuracy needs.
Is it license-safe for commercial use?
Yes — everything we serve today (Boltz-2, OpenFold-3, BoltzGen) permits commercial use with the required attribution, and we vet future additions the same way before enabling them.
How do I use BoltzGen for binder design?
BoltzGen requires a target structure (PDB/CIF URL from RCSB or similar). Provide the target URL and sequence length (which determines binder size). BoltzGen will design novel proteins that bind to your target structure.
Can I pick the model?
Yes. Use mode=auto or pin a backend explicitly.
Where does my data live?
EU by default, GDPR-compliant. Zero-retention mode available.
For investors
We're standardizing structure prediction into a developer-first primitive: a commercial‑use, EU-native folding substrate that routes across models and returns publishable outputs. An API for developers and a web UI for researchers.