Private AI Email Sorting: What Data Stays Local

The problem: you want AI sorting, but not a black box

Semantic email sorting needs an AI model to actually read your mail — that’s not optional, it’s how it works. The part that’s optional is whether you know what that means for your data. Most rivals in this space are cloud-only, black-box AI — they don’t tell you what gets sent where or how long it’s kept, so you’re left with the same open question every time: what actually happens to your mail once it’s been classified? If you’ve held off on AI email tools specifically because you couldn’t get a straight answer to “what actually leaves my machine,” that caution is reasonable — the fix is a tool that answers it precisely instead of a tool you have to trust blindly.

Why classic filters don’t solve this (they just avoid the question)

Sieve rules and provider-side filters are private by default because they’re dumb — matching a sender address or subject keyword happens locally and never needs to understand meaning. That privacy is a side effect of the filter being unable to do semantic sorting at all, not a design choice. The moment you want a filter that understands “this is a sales pitch” versus “this is a colleague,” something has to evaluate meaning, and that something is an LLM. The real question isn’t whether to involve an LLM — it’s exactly what you send it and what happens to that data afterward.

What actually happens with your data

Here’s the precise answer, not a vague privacy policy summary. For classification, AI Email Filter sends the subject, sender, and a short body preview to the LLM — not the full email. Long-term storage keeps only headline data: subject, sender, date, and the chosen category, so the History tab can show what happened and let you undo it. The full email body is never persisted anywhere.

A couple of the actual category prompts show how little context classification needs in practice:

• “From people I know personally or have emailed before” — a decision made from sender and prior contact, not from reading the message content in depth.
• “Cold outreach, LinkedIn spam, or someone trying to sell me something” — identifiable from subject and a preview, without needing the full body text.

If even that short preview going to a third-party API is more than you want, Bring Your Own LLM lets you point classification at a local Ollama instance instead. With a local model, no email content leaves your own infrastructure at all — classification happens on hardware you control, not a vendor’s servers.

Where the audit trail comes from

The reason anything is stored long-term at all is History and one-click Undo — every move, flag, forward, or delete is logged so you can see what the AI did and reverse it. That log only needs the headline data (what happened, when, to which message) to work; it doesn’t need the email body, which is exactly why the body isn’t kept. Nothing AI Email Filter does is permanent or silent — every action is visible and reversible from that log.

FAQ

What exact data gets sent to the LLM when an email is classified?

Subject, sender, and a short body preview — not the full email content. That’s the minimum needed for the category prompts to evaluate what an email is about.

Is the full email body stored anywhere after classification?

No. Long-term storage is limited to headline data — subject, sender, date, and the assigned category — kept so the History tab can display and undo past actions. The full body is never persisted.

Can I avoid sending anything to a third-party AI API at all?

Yes, with Bring Your Own LLM pointed at a local Ollama instance. In that setup, classification runs on your own hardware and no email content leaves your infrastructure.

Try it with full visibility into what’s sent

AI Email Filter includes 200 free classifications, no credit card required, so you can see exactly what data moves before deciding whether to trust it with your real inbox. See pricing for plans beyond the free tier.