Boltz-2
Predicts a protein’s 3D structure from its sequence.
- Runs on
- Modal · USA
- Hardware
- H100 GPU
Salpa Compute · Free tier
AI models as nodes in your workflow, on hardware we run
Your workflow stays on your machine; its cloud nodes run on ours. Five open models for structure prediction, docking, protein design and genomics, four of them on GPUs, with a free tier: no cloud account, no setup.
Five models · one GPU-hour a month free · no cloud account
Local workflow, remote hardware
One workflow, run in two places
Build and run your workflow on your own machine, as you do today. Add one of the cloud nodes and only that step runs on Salpa Compute; its results come back into your workflow for the next node.
- Your node on your machine
- Cloud node on Salpa Compute
- Your node on your machine
The models
Five open models, each a node you place in a workflow.
Chai-1
Predicts the 3D structure of complexes: proteins with ligands or nucleic acids.
- Runs on
- Modal · USA
- Hardware
- H100 GPU
DiffDock
Blind protein–ligand docking: finds binding poses without a predefined pocket.
- Runs on
- Modal · USA
- Hardware
- A10G GPU
Evo2
A DNA foundation model: score, embed or generate genomic sequences.
- Runs on
- Google Cloud · Belgium
- Hardware
- L4 GPU
ProteinMPNN
Inverse folding: designs amino-acid sequences that fold to a given backbone.
- Runs on
- Google Cloud · USA
- Hardware
- CPU, no GPU needed
A model you need?
Tell us which model or tool you would like as a cloud node.
Getting started
No waitlist and no approval: sign in and run.
Sign in from Salpa
With Google or an email address. No cloud account of your own, no waitlist.
The free tier is set up
The first time you run a cloud node, your account is created on the free tier.
Add a cloud node
Place it in your workflow like any other node. It runs on our hardware.
Results come back
Its outputs return into your workflow, on your machine, for the next node to use.
The free tier
The terms ↗- Allowance
- One GPU-hour a month
- At a time
- Two GPU jobs in progress
- Beyond it
- Jobs are refused, never billed
- More capacity
- By arrangement: write to us
- Your data
- Kept only for the job
A large result waits at most 24 hours for you to download it. Nothing you send is used to train models. If your data should not leave the EU, keep to local nodes: every cloud node is optional. Where each model runs, and what is kept for how long: the privacy notice ↗
Powered by Modal and Google Cloud
Who also support us through their startup programmes.
Bring cloud nodes into your workflow.
Start with Salpa, sign in, and add a cloud node to your workflow.