Only on your click
No mail is sent to an AI model automatically. Nothing runs in the background, nothing gets analyzed just in case. Without your click, nothing happens.
Privacy promise
Most AI mail tools copy your mailbox to their servers so they can search it fast. PixMail doesn't. This page explains what happens instead, and where the limits are.
In short
They're built into the architecture, not just the marketing.
In black and white
The left column is complete. If something you care about is missing from it, ask, and we'll add it here.
The path of a mail
When you open a mail, here's what happens:
The interface runs on your device. It asks our API for one specific message from one specific folder.
It signs in to your mail server with your decrypted credentials, fetches exactly that one message, and sends it on to your browser. Nothing gets written along the way.
Your messages stay right where they were, with your provider. Delete PixMail, and your mailbox is untouched.
That's why PixMail checks in briefly the first time you open a large folder: it genuinely asks again each time, instead of answering from a copy.
AI tools
The AI doesn't run on our servers, it runs at Anthropic or OpenAI. That's why what goes there matters.
No mail is sent to an AI model automatically. Nothing runs in the background, nothing gets analyzed just in case. Without your click, nothing happens.
For text AI, recognised personal details in the selected content are replaced with placeholders before transmission and restored in the answer. Details in free text may go undetected. Your password and entire mailbox are not transmitted.
We don't store the reply. What lands in your usage log afterward is the tool, the model, the token count, and the price. The mail content itself doesn't.
Staying honest
A promise that only lists the good parts isn't one.
Anthropic's first-party API has no EU region. If an action runs through Anthropic, the text is processed in the US. With OpenAI, processing in Europe is possible. Which provider serves which tier, you can see in the app before every action.
Both keep requests for a while to check for abuse, usually seven days at Anthropic, up to 30 at OpenAI. Under both providers' terms, the data isn't used to train their models. We don't verify that ourselves, we just pass on what's promised there.
PixMail is the interface in front of it. How securely your mail is stored, who has access there, and how long deleted messages stick around is up to the provider that holds your mailbox.
The privacy policy says the same thing again, in the language the law requires, with legal grounds, retention periods, and your rights.