Your data, secured in your hands.
Most AI services become more valuable the more of your data they hold. Ours does not. We train on it once, on hardware we own, and give you the result.
Our commitments
Isolated training hardware
Training runs on GPU machines we own and operate, not a shared cloud training pool. Your job runs alone on its machine.
No third-party training services by default
Your data is not sent to an external fine-tuning API unless you ask for a specific provider in writing.
Used for your model only
Your data trains your model and nothing else. It is never pooled, never used to improve a shared model, never sold.
Deleted on delivery
When you accept the delivery, training data, intermediate checkpoints and logs are deleted. You can request a deletion record.
You own the weights
The fine-tuned weights, adapter and eval report are yours. Standard formats, no licence back to us, no lock-in.
Hosted where you choose
We do not run your model in production. It runs in your AWS account, your Cloudera cluster, your servers or your laptop.
What we need, and what we do not
We need enough examples of the task to teach it and to test it. We do not need live access to your systems, your production database or your customer accounts. Export what is relevant, remove what is not, and send it through the encrypted channel we agree at intake.
Personal data
Before training we scrub personal data that the task does not need. Where the task needs it (for example a model that reads invoices must see supplier names), we say so in the intake plan and you decide.
Traceability
Every delivery includes a training manifest: base model and licence, data hash and row count, training settings, the machine used, who approved the eval. It answers the questions a security review will ask.
Options for regulated teams
- On-site training on your own hardware, with our tooling.
- Signed data processing agreement.
- Key-by-key review of the deploy recipe before anything touches your environment.
Short version: we are a tailor, not a landlord. We cut the model to fit you and hand it over. We keep nothing.