AI infrastructure
for Australian research
GPU compute, model hosting, and training jobs — with data that stays in Australia. One API, from laptop to cluster.
Built for sovereign compute
Yarn is a GPU-aware AI platform designed from the ground up for Australian data residency, research workflows, and elastic scaling.
Sovereign by default
Every request, every token, every byte of training data stays on Australian soil. No data leaves. No exceptions.
GPU compute on demand
RTX 4090 on bare metal today. H100 cloud spokes when you need them. One API, any scale.
Research-first platform
Submit training jobs, launch interactive GPU sessions, fine-tune models. Built for researchers, not DevOps engineers.
OpenAI-compatible API
Drop-in replacement. Switch one base URL and your existing code works. Streaming, function calling, embeddings.
Three lines to GPU
The Yarn SDK handles authentication, job submission, and result retrieval. Install it, point it at your code, and let the platform handle the rest.
OpenAI-compatible inference is even simpler — change the base URL and you're done.
Read the docs$ pip install yarn-au
import yarn
job = yarn.submit_job(
directory="./my-experiment",
gpu="rtx4090"
)
job.wait()
print(job.result)Simple, transparent pricing
Start free with your own GPU. Pay only for what you use with credit packs. Or talk to us about dedicated capacity for your team.
BYO GPU
Free
Bring your own hardware. Share compute across your team or institution.
- Connect your GPUs to the Yarn network
- Share compute between teams and researchers
- Manage distribution across an entire institution
- Australian data residency
- Overflow to shared pool when you need more
Managed compute
From $0.20/hr
Set clear budgets and plan your spending. Per-second billing, full cost visibility, no surprises.
- Training jobs and interactive GPU sessions
- Chat and API inference
- Budget limits, alerts, and auto-reload
- Per-team and per-user spend controls
- No subscriptions, no lock-in
Start building on sovereign infrastructure
Create a free account, connect your GPU or buy credits, and deploy your first model in minutes.