Workflows

Five open design engines, compared.

Ariax runs the original open-source protein-design methods, unmodified. This page summarizes what each one designs, the architecture behind it, who built it, and where its code lives, so you can pick the right engine before you start a campaign.

Published methods with public code
No separate software license to obtain
Pay only for the GPU time you use

Side by side

Architecture, origin, and license.

Release dates are the first public release of the method. Licenses are those of the upstream project; on Ariax there is never a separate software fee.

Protein design engines available on Ariax Bio, compared by model architecture, binder formats, supported targets, release date, creator, and license.
Engine	Model architecture	Designs	Targets	Released	Created by	License	Code
BindCraft2Default	AlphaFold2-guided design with modality presets	Fab, scFv variable domains, VHH / nanobody, Large binder, Miniprotein, Ankyrin repeat (ARP), Linear peptide, Cyclic peptide	Proteins, including multi-target and detargeting objectives	Sep 2026	Martin Pacesa and the BindCraft2 contributorsPacesa Lab, University of Zurich	Source-available	BindCraft2PacesaLab
BindCraft	AlphaFold2 hallucination + ProteinMPNN	Miniprotein, Linear peptide	Proteins	Sep 2024	Martin Pacesa, Bruno Correia and colleaguesLaboratory of Protein Design and Immunoengineering, EPFL	MIT	BindCraftmartinpacesa · original FreeBindCraftcytokineking · Ariax-maintained, PyRosetta-free
BoltzGen	All-atom generative model	VHH / nanobody, Cyclic peptide, Helicon (stapled helix), Miniprotein, Linear peptide	Proteins, nucleic acids, small molecules (CCD/SMILES), peptide hormones	Oct 2025	Hannes Stärk and the Boltz teamMassachusetts Institute of Technology	MIT	boltzgenHannesStark
PXDesign	Diffusion Transformer (DiT) + dual-filter ranking	Miniprotein	Proteins	Aug 2025	ByteDance AI for Science, the team behind ProtenixByteDance	Apache 2.0	PXDesignbytedance · original PXDesigncytokineking · Ariax cloud-optimized fork
ESMFold2-pipeline	ESMFold2 inversion + Protenix validation	Miniprotein, VHH / nanobody, scFv	Proteins	Jun 2026	Pipeline by Ariax Bio; ESMFold2 by BiohubAriax Bio, on Biohub's ESMFold2	Elastic License 2.0	esmfold2-pipelinecytokineking

Licenses reflect the upstream projects at the time of writing; check each repository for current terms. BindCraft2 is source-available and restricts third-party hosting — Ariax holds the license needed to offer it as a hosted workflow.

Binder formats

Which engine designs what.

Start from the format the biology calls for, then read across to see which engines support it. Several formats are available from more than one engine.

Binder format support by engine. A check mark means the engine designs that format on Ariax Bio.
Format	BindCraft2	BindCraft	BoltzGen	PXDesign	ESMFold2-pipeline
Miniprotein	Supported	Supported	Supported	Supported	Supported
Linear peptide	Supported	Supported	Supported	Not supported	Not supported
Cyclic peptide	Supported	Not supported	Supported	Not supported	Not supported
Helicon (stapled helix)	Not supported	Not supported	Supported	Not supported	Not supported
VHH / nanobody	Supported	Not supported	Supported	Not supported	Supported
scFv	Not supported	Not supported	Not supported	Not supported	Supported
Fab	Supported	Not supported	Not supported	Not supported	Not supported
Ankyrin repeat (ARP)	Supported	Not supported	Not supported	Not supported	Not supported
Large binder	Supported	Not supported	Not supported	Not supported	Not supported

In detail

How each engine works.

BindCraft2Default

The BindCraft approach extended across therapeutic formats.

Architecture

Keeps the core BindCraft strategy — optimize through AlphaFold2, redesign with ProteinMPNN, re-predict and filter — inside a broader system of format-specific presets and biological objectives, on a much faster campaign engine.

Background

BindCraft2 is the successor to BindCraft from the same lead author. It widens the original method from miniproteins to antibody fragments, repeat-protein scaffolds and cyclic peptides, and adds specialized homo-oligomer and multidomain setups. It is now our default starting point for most campaigns.

Best for

Our default recommendation, including for miniproteins. Also the widest range of binder formats, scaffolded antibody and ARP design, and campaigns with multi-target or conformational objectives.

Created by: Martin Pacesa and the BindCraft2 contributorsPacesa Lab, University of Zurich
Released: Sep 2026Repository published September 2026; available on Ariax from 21 September 2026.
License: Source-availableThe BindCraft2 Source-Available License restricts third-party hosting. Ariax holds the licensing needed to offer the hosted workflow, so you pay only for compute with no separate commercial license to obtain.
Code: PacesaLab/BindCraft2
On Ariax: Setup guide Read the announcement

BindCraft

Automated de novo miniprotein and peptide design.

Architecture

Optimizes a binder by backpropagating through AlphaFold2 (via ColabDesign), redesigns the sequence with ProteinMPNN, then re-predicts and filters the result. On Ariax you choose PyRosetta or the fully open-source FreeBindCraft scoring path.

Background

BindCraft is the method that inspired Ariax. Its Nature study reported experimental success rates of 10–100% across the targets it tested, and it has held up in independent competitions. BindCraft2 has since become our default recommendation, but the original remains a proven, highly automated option.

Best for

The original, battle-tested miniprotein and linear-peptide workflow, or fully open-source scoring through FreeBindCraft.

Created by: Martin Pacesa, Bruno Correia and colleaguesLaboratory of Protein Design and Immunoengineering, EPFL
Released: Sep 2024Preprint September 2024; published in Nature on 27 August 2025.
License: MITMIT for the pipeline itself. Its default scoring step uses PyRosetta, which carries a separate Rosetta license; FreeBindCraft replaces that step with open-source tools.
Code: martinpacesa/BindCraftoriginal
cytokineking/FreeBindCraftAriax-maintained, PyRosetta-free
Paper: Nature (2025)
On Ariax: Setup guide Read the announcement

BoltzGen

All-atom generative design across binder and target classes.

Architecture

An all-atom framework from the team behind the Boltz structure-prediction models, with a rich binding-site constraint language. Because it reasons over all atoms, it can design against target classes beyond proteins.

Background

BoltzGen brought open-source design to modalities that had resisted it, notably VHH single-domain antibodies. The team validated it on targets with no close relatives in the Protein Data Bank, so the model could not fall back on memorized binding modes.

Best for

VHH nanobodies, cyclic peptides and helicons, or any campaign that needs all-atom reasoning or a non-protein target.

Created by: Hannes Stärk and the Boltz teamMassachusetts Institute of Technology
Released: Oct 2025Repository published October 2025; preprint posted November 2025.
License: MITFully open source under the MIT license for academic and commercial use.
Code: HannesStark/boltzgen
Paper: bioRxiv (2025)
On Ariax: Setup guide Read the announcement

PXDesign

Diffusion-based miniprotein design with dual-filter ranking.

Architecture

Pairs a diffusion-based backbone generator with a ranking stage that filters on both Protenix and AF2-IG confidence. Generating Cartesian coordinates directly avoids the expensive equivariant operations used by older diffusion models.

Background

PXDesign comes from the group that released Protenix, an open reproduction of AlphaFold3. Having a diffusion engine alongside hallucination-based BindCraft matters in practice: different architectures succeed on different targets, so running both raises the odds on a hard one.

Best for

A diffusion-based alternative to BindCraft for miniproteins, with greater structural diversity and confidence scores from two independent models.

Created by: ByteDance AI for Science, the team behind ProtenixByteDance
Released: Aug 2025Preprint posted August 2025; repository published December 2025.
License: Apache 2.0Fully open source under Apache 2.0, free for academic research and commercial use.
Code: bytedance/PXDesignoriginal
cytokineking/PXDesignAriax cloud-optimized fork
Paper: bioRxiv (2025)
On Ariax: Setup guide Read the announcement

ESMFold2-pipeline

Campaign-scale miniprotein, VHH and scFv design.

Architecture

Optimizes candidate sequences against the structures and interfaces predicted by ESMFold2, which combines the ESMC-6B protein language model with a recurrent pair-folding architecture and an all-atom diffusion model. Every shortlist is validated independently with Protenix v2 and ranked by consensus.

Background

Biohub published ESMFold2 with a binder-design tutorial. Ariax turned that method into a production campaign system with parallel design, checkpointing and shortlist generation. Ariax is the only hosted platform where it runs, and the pipeline is not licensed to other providers.

Best for

scFv campaigns, and miniprotein or VHH work where you want mandatory independent validation and consensus ranking.

Created by: Pipeline by Ariax Bio; ESMFold2 by BiohubAriax Bio, on Biohub's ESMFold2
Released: Jun 2026ESMFold2 released by Biohub in May 2026; the Ariax pipeline followed in June 2026.
License: Elastic License 2.0The pipeline is source-available under Elastic License 2.0, which permits royalty-free commercial use on your own infrastructure. ESMFold2 code and weights come from Biohub.
Code: cytokineking/esmfold2-pipeline
Paper: bioRxiv (2026)
On Ariax: Setup guide Read the announcement

Start a campaign

Pick an engine and run it.

Every method on this page runs on managed GPUs with no software fee. Explore the platform for free and pay only for the compute your designs use.

Get Started Free View Pricing

Billing: Per GPU-hour
Subscription: None
Queue: None
Workflows: 5 methods

Engine

Model architecture

Designs

Targets

Released

Created by

License

Code

BindCraft2Default

AlphaFold2-guided design with modality presets

Fab, scFv variable domains, VHH / nanobody, Large binder, Miniprotein, Ankyrin repeat (ARP), Linear peptide, Cyclic peptide

Proteins, including multi-target and detargeting objectives

Sep 2026

Martin Pacesa and the BindCraft2 contributorsPacesa Lab, University of Zurich

Source-available

BindCraft2PacesaLab

BindCraft

AlphaFold2 hallucination + ProteinMPNN

Miniprotein, Linear peptide

Proteins

Sep 2024

Martin Pacesa, Bruno Correia and colleaguesLaboratory of Protein Design and Immunoengineering, EPFL

MIT

BindCraftmartinpacesa · original
FreeBindCraftcytokineking · Ariax-maintained, PyRosetta-free

BoltzGen

All-atom generative model

VHH / nanobody, Cyclic peptide, Helicon (stapled helix), Miniprotein, Linear peptide

Proteins, nucleic acids, small molecules (CCD/SMILES), peptide hormones

Oct 2025

Hannes Stärk and the Boltz teamMassachusetts Institute of Technology

MIT

boltzgenHannesStark

PXDesign

Diffusion Transformer (DiT) + dual-filter ranking

Miniprotein

Proteins

Aug 2025

ByteDance AI for Science, the team behind ProtenixByteDance

Apache 2.0

PXDesignbytedance · original
PXDesigncytokineking · Ariax cloud-optimized fork

ESMFold2-pipeline

ESMFold2 inversion + Protenix validation

Miniprotein, VHH / nanobody, scFv

Proteins

Jun 2026

Pipeline by Ariax Bio; ESMFold2 by BiohubAriax Bio, on Biohub's ESMFold2

Elastic License 2.0

esmfold2-pipelinecytokineking

Format

BindCraft2

BindCraft

BoltzGen

PXDesign

ESMFold2-pipeline

Miniprotein

Linear peptide

Cyclic peptide

Helicon (stapled helix)

VHH / nanobody

scFv

Fab

Ankyrin repeat (ARP)

Large binder

Architecture

Background

Best for

Our default recommendation, including for miniproteins. Also the widest range of binder formats, scaffolded antibody and ARP design, and campaigns with multi-target or conformational objectives.

Created by: Martin Pacesa and the BindCraft2 contributorsPacesa Lab, University of Zurich
Released: Sep 2026Repository published September 2026; available on Ariax from 21 September 2026.
License: Source-availableThe BindCraft2 Source-Available License restricts third-party hosting. Ariax holds the licensing needed to offer the hosted workflow, so you pay only for compute with no separate commercial license to obtain.
Code: PacesaLab/BindCraft2
On Ariax: Setup guide Read the announcement

Architecture

Background

Best for

The original, battle-tested miniprotein and linear-peptide workflow, or fully open-source scoring through FreeBindCraft.

Created by: Martin Pacesa, Bruno Correia and colleaguesLaboratory of Protein Design and Immunoengineering, EPFL
Released: Sep 2024Preprint September 2024; published in Nature on 27 August 2025.
License: MITMIT for the pipeline itself. Its default scoring step uses PyRosetta, which carries a separate Rosetta license; FreeBindCraft replaces that step with open-source tools.
Code: martinpacesa/BindCraftoriginal
cytokineking/FreeBindCraftAriax-maintained, PyRosetta-free
Paper: Nature (2025)
On Ariax: Setup guide Read the announcement

Architecture

Background

Best for

VHH nanobodies, cyclic peptides and helicons, or any campaign that needs all-atom reasoning or a non-protein target.

Created by: Hannes Stärk and the Boltz teamMassachusetts Institute of Technology
Released: Oct 2025Repository published October 2025; preprint posted November 2025.
License: MITFully open source under the MIT license for academic and commercial use.
Code: HannesStark/boltzgen
Paper: bioRxiv (2025)
On Ariax: Setup guide Read the announcement