Writing
Emails, posts, flyers, proposals, the reply you have rewritten three times.
Now in private beta across the United States
Private Safe AI is an AI assistant for writing, planning, learning and the small-office jobs that eat a morning. You ask, it answers, and the conversation stays on the device you had it on.
Nothing to pay during the beta.
Tell us your city when you write.
I have three bills due this week and a busy Thursday. Help me plan it.
Pay the two online ones tonight, they take five minutes. Mail the third on Wednesday so Thursday stays clear. Want a short checklist?
Yes, a checklist please.
Your conversations are saved only on your device, in your browser’s storage. Nothing is saved on our servers. Here is the whole trip a message takes, with nothing left out.
You write a message. The conversation is saved here, and only here. Use the one-tap export to keep a backup.
Passes your message to our AI provider and back. Its logs contain only your account ID, token counts and cost. Never message content.
Reads your words briefly, in memory, to write the reply. For the models we serve, its published policy is to delete them after and never train on them.
Everyday work, minus the feeling that someone is keeping a file on you.
Emails, posts, flyers, proposals, the reply you have rewritten three times.
Lists, schedules, checklists, the agenda for a meeting nobody has written yet.
A term, a document or a decision explained in plain language, at your pace.
Scheduling messages, reminders, marketing drafts, bookkeeping summaries, inventory lists. A person still approves anything before it leaves.
We would rather say this here than have you find out later.
To write a reply, the provider’s servers must see your words briefly, in memory.
Anyone with access to your unlocked device can read your chats. Use your device lock.
It is not a lawyer, doctor or financial advisor, and not HIPAA compliant. In an emergency, call 911.
The open models behind Private Safe AI (the DeepSeek, GLM and MiMo families) run on servers in the United States operated by DeepInfra, a US company. The GLM models are published by Z.ai, a company based in China; your conversations are processed in the United States and are not sent to Z.ai or to any other AI model developer.
We never train on your conversations. For the open-weight models we serve, our provider’s published policy is that inputs and outputs are held in memory only while the reply is generated, then deleted, and API data is never used for training. We serve only models whose providers don’t store or train on your data.
Request beta access by email and tell us your city or ZIP code. A person reads every request, and we open spots in small groups. Nothing to pay during the beta.
Private Safe AI is a Tufts Media LLC company. It runs the same private AI as Glass City AI, our local service in Toledo, Ohio, for people anywhere in the United States.
We don’t send marketing email today. No newsletter, no mailing list, nothing scheduled. To reach us, write [email protected]. Ask for a human and you’ll get one.
Private Safe AI began in Toledo, Ohio, as Glass City AI, where we help people learn AI in person. Same private AI, now open to the rest of the country. In the Toledo area? Glass City AI is your local door.
Private beta. A person reads every request.