A human in the loop, by architecture
In työ, a draft is just a draft. Every AI-suggested reply is held until a person approves it — nothing is sent automatically, ever. You read the draft, change what you want, and send it as your own message.
For sensitive replies — refunds, complaints, anything with money or feelings in it — you can require sign-off from more than one person before the message goes out.
Every approval leaves a trail: who approved which message, and when. If you ever need to show how a reply got sent, the record is there.
Your data stays in the EU
Workspace data is stored and processed in the EU. The AI drafting runs on EU-hosted models too — your customers' messages aren't shipped outside the EU to get a draft back.
GDPR-aligned customer data controls
työ remembers useful things about your customers — their language, their usual order, an allergy they mentioned — so replies get better over time. You can see exactly what it remembers about any customer in the customer memory dashboard, and delete it. Memory you can't inspect isn't a feature; it's a liability.
What we promise — and what we don't
We don't promise that työ makes you compliant with the AI Act. No tool can, because compliance depends on your whole business, not on one piece of software.
What we do promise: työ is designed for the human-oversight approach the AI Act rewards. Your customers always talk to a person. AI assists, a human decides, and the record keeps everyone honest.