Data
Client data stays in your systems unless the engagement says otherwise. We do not use your data to train public models. Access is scoped to the work, revoked at the end, and logged.
How we handle the work and the data.
This page is operating practice. It is not a certificate. Where we do not have a public attestation, we say so.
Models
We are not a reseller tied to one lab. Model choice is for the job, written into the engagement, and changeable. If a vendor term would send your data somewhere you have not agreed, we do not use that path.
Subprocessors
Infrastructure is typically AWS when that is the stack. Any other processor is named in the engagement before it sees data. There is no hidden list on a marketing page.
Retention
Working copies last for the engagement plus a short wind-down we agree in writing. We do not keep a private warehouse of client corpora.
Contracts
Data terms go in the engagement contract. A DPA can be part of that when the work needs it. We have not published a SOC 2 report on this site; we will not pretend otherwise.
High-stakes actions
Autonomous steps that spend money, change customer records, or cannot be undone keep a human approval path unless you explicitly take that off. That is how we run work — not a legal opinion on the EU AI Act.
Entity
Arcift Technologies. Contact: [email protected]. Remote team, reachable on North American hours. Incorporation jurisdiction is confirmed in the contract, not invented here.