Is it safe to connect AI to my email and Slack?
It is when the limits are enforced by the platform, not by the prompt. In Kazi each connection is set to Only read or Can update, and every call is checked against that before it runs. We don't train our models on your content.
Ken Wiltshire, founder of Kazi · Checked against the product on 2026-11-16
Start with read-only
For a morning rundown, Kazi only needs to read. Set email and Slack to Only read and it can't send or post, whatever the model decides. Open up Can update on one app when you have a job that needs it.
With Ask Mode on, which is how every new web tab starts and how voice always works, a change through an Only read connection stops and asks you first. Background agents just get refused.
What else to check
Who can see what. In a team, each project's agent has its own access list per app, so the sales agent doesn't read the finance inbox.
Where your data goes. We don't use your content to train our own models. The model providers that run your requests have their own data practices, and for regulated firms there's Kazi Private.