No AI subprocessor
The model ships inside the app and runs on the user's machine. Nothing is sent to OpenAI, Anthropic, Google, or to us. There is no third party to add to your register.
Enterprise
AI email for teams that can't send client data to a vendor.
Every other AI email tool works by uploading your inbox. For firms holding privileged, regulated or commercially sensitive mail, that is not a procurement question. It is a non-starter. TurboMail runs the model on each machine instead.
The difference
There is no vendor to trust, because we never receive the mail.
Nothing to train on
We cannot train on your email for the same reason we cannot read it. The question most AI vendors answer with a policy, this answers with architecture.
Credentials stay local
Accounts connect through Google's and Microsoft's own sign-in. Tokens are encrypted on the device. Nothing authenticates from our infrastructure.
In practice
Where mail livesCached on the device, encrypted at rest. It is never copied to our servers, so a breach of ours cannot expose it.
Where AI runsOn the machine, using a bundled local model. It works offline, on a plane, and inside a closed network.
Cloud, if you want itTeams that prefer a frontier model can add their own API key. The traffic goes from the machine straight to the provider you chose, under your contract with them, and the app always shows which model is active.
What reaches usAnonymous install and active-user counts, keyed to a random id. No addresses, subjects or content, ever.
PlatformsmacOS and Windows desktop today. Mobile is on the roadmap, not shipped.
Get in touch
Tell us about your team.
A person reads every one of these, usually within a day.