Features

AI on your key, or on no key at all

There is no MailDock inference service. Whatever the AI features do, they do with a provider you chose and pay, or with a model running on your own hardware.

See pricing What leaves the server

The admin console’s AI section: choosing a provider, its endpoint and model, and a switch that keeps every request on the local network.

In the mailbox

Thread summaries, an explain-this-email that uses the spam verdict rather than guessing, smart replies that are never sent automatically, and compose assist for drafting, rewriting, translating and fixing grammar.

Thread summary
Explain this email
Smart replies — suggested, never sent
Draft, rewrite, translate, fix grammar

In the console

An admin copilot with read-only tools. It can look at the estate and propose a change; the change is a stored proposal you approve by id, so the model never holds a write.

Read-only tools
Write actions are proposals approved by id

What leaves the server

Per-user opt-in under a per-domain policy, so an administrator decides what is available and a user decides whether to use it. A user opts in under Settings → AI features in webmail, and the switch appears only where the domain allows AI. Attachments are excluded unless an administrator allows them for the domain, and no feature reads one yet. The call log records metadata, not content; the admin copilot's tool log keeps the arguments it ran with, on your server. Keys are encrypted at rest. Set localOnly and nothing leaves at all.

Per-user opt-in, per-domain policy
Attachments excluded
Metadata-only audit
Keys encrypted at rest
localOnly with Ollama or vLLM

Providers

Anthropic, OpenAI, Gemini, xAI, Meta Llama and Groq, plus vLLM and Ollama for a model you host yourself.

Where these claims come from

Each line on this page is backed by a document in the product’s own repository. They are private, so the paths are given rather than linked; ask us if you want to see one.