Your environment.
A defined run.
A customer-operated runner picks up an approved job. Models, data and compute stay within the configured execution boundary.
Execution, with a record.
Your infrastructure.
Your models.
A verifiable record.
Keep supervised ML runs on infrastructure you operate. Carry the signed execution evidence into your next review.
Research preview · Supervised execution · Customer-operated runners
01 / The evidence layer
Compute stays with you.Evidence travels.
A smaller layer. A clearer record.
From a supervised run
to something you can inspect.
CloudTune connects execution, recovery and evidence around the infrastructure you already operate.
Your environment.
When execution stops,
ownership matters.
Leases and generation fencing separate the current owner from an expired attempt. Late results cannot overwrite a newer generation.
A record you can take.
A signature you can check.
Export signed evidence and verify it offline against a trusted runner key. The record remains inspectable when the control plane is unavailable.
A signature establishes integrity and signer identity against a trusted key. It does not prove model quality or that training occurred.
Touch the evidence.
Make a record.
Then try to break it.
Generate a signed example, change its contents, and watch verification reject it. The trust anchor is kept separate from the receipt.
CloudTune / evidence labLOCAL TO YOUR BROWSER
Simulated execution.
Real browser cryptography.
This example runs no training and makes no API request. It uses an illustration schema, separate from production runner receipts.
Current step
Ready to create a demo receipt
The signing key is created in memory for this demonstration. This is local demonstration trust, not a production identity.
receipt.jsonEd25519 + SHA-256
{
"kind": "cloudtune-browser-demo/v1",
"state": "waiting_for_your_first_receipt"
}Nothing to install. No dataset to upload. The receipt is generated on this device.
The current boundary
Clear about what exists.
Clear about what is next.
This is a research preview with reviewed engineering evidence; production qualification is not ready. External validation remains open.
Supervised ML execution
- Supervised QLoRA on approved Qwen2.5 0.5B and 1.5B configurations
- Customer-operated runners and explicit job ownership
- Generation fencing, recovery controls and portable signed receipts
- Offline verification against a trusted runner key
Not production-qualified
- Engineering handoff and regression evidence have been reviewed
- An independent colleague execution remains pending
- Customer workflow validation remains pending
- No claim of certified compliance, paid adoption or enterprise readiness
Start with the workflow.
What would your
reviewer need to see?
Bring a recent ML run, the review that followed, and the evidence that was missing. That is the conversation CloudTune needs next.
Talk to the builder Yanze Wu on LinkedIn (opens in a new tab)Founder-direct. A workflow conversation.