Guide · Privacy

A self-hosted alternative to cloud AI inbox assistants

Updated 3 November 2026

AI assistants for your inbox are useful: they sort, summarise and draft. Most of them work the same way, though. You connect your mail account, and the service reads and stores your mail on its own servers. For many people that also means family chats, receipts, health and money matters, and everything their colleagues write to them.

Before you connect any assistant, it's worth asking five questions.

Five questions to ask any AI inbox assistant

  1. Where are my messages stored? On your devices, or as a copy on the provider's servers?
  2. Who processes them, and under which terms? Which AI provider reads them, and may your data be used for training?
  3. What can it do on its own? Can it send, delete or move mail without you confirming it?
  4. Can I see what it does? Is there a log of every run and every action, and can you read the code?
  5. What happens when I stop? Is the copy deleted, and can you prove it?

What "self-hosted" changes

A self-hosted assistant runs on a machine you own. With Axella, that's a Mac that stays on at home, often a Mac mini:

What still leaves your home – said plainly

Self-hosted doesn't mean nothing leaves. Today, Axella uses Claude, on your own Anthropic subscription, to read what each digest needs and to write the drafts. That text is processed on Anthropic's servers under your account's terms. Topic briefs send only their search queries, and the optional call agent runs on Twilio and Vercel. Fully local processing, with a model running on the Mac, is the goal as local models get good enough.

The difference from a typical cloud assistant: no third party keeps a running copy of your inbox, and you decide which parts are used at all.

The trade-offs

See what it costs and what you need, try the demo with example data, or go straight to the setup guide.

More guides