> ## Documentation Index
> Fetch the complete documentation index at: https://docs.langmail.me/llms.txt
> Use this file to discover all available pages before exploring further.

# Sovereignty

> What sovereign means mechanically at Langmail — your own mail server, open protocols as the right to leave, and mail that is never training signal.

"Sovereign" is a mechanism at Langmail, not a policy promise. Most claims below are architectural properties you can verify from the outside; where a claim is a commitment rather than a mechanism, we say so.

## No hyperscalers

Your mailbox lives on Langmail's own mail server on European infrastructure, under GDPR jurisdiction. There is no Google or Microsoft account underneath — `mail.langmail.me` is the storage of record, and Langmail operates it end to end.

Connected Gmail accounts are the exception you control: if you sync a Gmail account in, that mail obviously also exists at Google. Your native `@langmail.me` mail never touches a hyperscaler.

## Open protocols are the right to leave

Sovereignty is only real if leaving is cheap. Every layer of Langmail speaks a standard anyone can implement:

| Layer        | Protocol          | What it means for you                                                     |
| ------------ | ----------------- | ------------------------------------------------------------------------- |
| Mail         | JMAP, IMAP, SMTP  | Any standard client or export tool reads your entire archive              |
| Calendar     | CalDAV, iCalendar | Your events open in Apple Calendar, Thunderbird, or a plain `.ics` file   |
| Agent access | MCP, OAuth 2.1    | Your agent integration isn't proprietary either — any MCP client connects |

No proprietary formats, no export gatekeeping. The endpoints are documented in [Open protocols](/reference/protocols) — the same docs a competitor's migration tool would use.

## Your mail is not training signal

This one is a commitment, not a mechanism you can inspect: your mail is not used to train models. Langmail's classification and summaries work *for* your mailbox, not on it as a corpus. What makes the commitment credible is structural — Langmail's business does not depend on reading your mail, while an ad-funded incumbent's does.

## Agents without surrendered credentials

Agent access is designed so you never hand raw credentials to software you don't fully trust:

* Each MCP client gets its own OAuth grant, approved by you on a consent screen.
* Your agent authenticates with a token that resolves only to your mailbox — never a password, never account-wide credentials.
* Mail content is delivered to agents as data to reason over. The email tool surface is read-only, so a message can't trick your agent into sending, deleting, or altering mail — those tools don't exist. Calendar tools do send invitation emails to attendees the agent names; that is the one outbound path, and every tool result states when it was used.

## What we don't claim

Honesty is part of the register. Langmail today is a hosted service — you trust Langmail's operations rather than a hyperscaler's, and the codebase is not currently open source. What makes the trust cheap to withdraw is the open data layer: your mail is one IMAP sync away from anywhere else.