Agents & automation
Give an AI agent scoped access to your inbox — without ever handing over your keys.
MailGate can hand off inbox work to an agent — an automated helper that acts on your mailbox within limits you set. You grant it a few specific permissions, never your password or your keys, and everything it does is logged.
This is the most experimental part of MailGate, so expect some rough edges.
What an agent can do
An agent only ever has the permissions you give it. Grant the minimum it needs for the job:
- Read — read your mail.
- Send — send mail on your behalf.
- Gating — manage your whitelist and blacklist.
- Toll — adjust toll settings and pay a toll on your behalf.
An agent with only Read can't send a message or change a setting, no matter what it's asked to do. Permissions are enforced strictly — that's the whole point.
Setting one up
From the Agents page, add an agent, give it a name, and choose its permission scope. Its key is generated and held for you inside the secure enclave — you never handle it directly, and you can revoke the agent at any time.
Skills: what the agent actually does
Skills aren't available in the dashboard yet. The skill engine exists on the backend, but there's no UI to install or configure a skill from the Agents page today. Everything else on this page — creating an agent, setting permissions, the audit log, and connecting your AI tools — works now. This section describes what's coming.
An agent doesn't do anything on its own — you install a skill, a packaged capability it runs for you. Two are planned to ship first:
- Auto-reply — drafts replies to incoming mail in a tone and with rules you set. You choose whether it sends automatically or holds each draft for your approval.
- Triage — reads incoming mail and sorts it: categorizes messages, flags what looks urgent, and can hand you a short summary of what came in.
A skill will run when new mail arrives, on a schedule you set, or on demand when you trigger it yourself.
Approvals & the audit log
Once skills land, reading and organizing your mail will just happen — a skill like Triage never needs to ask. It's the actions that send mail or change a setting that pause for your sign-off. When a skill wants to do something you've asked it to confirm first — like sending a reply while auto-send is off — it shows up in the Approvals tab, where you can approve it, edit the draft before approving, or reject it.
Pending approvals expire after 24 hours. If you don't act on one, it disappears and the agent has to ask again the next time it runs.
Every action an agent takes is written to an audit log on the agent's detail view, so you can always see exactly what it did and when.
Good to know / current limits
- Skills aren't in the dashboard yet. Creating agents, permissions, and the audit log work today; installing and configuring skills is still on the way (see the note above).
- Skills will be first-party only at first. There's no third-party skill marketplace planned for launch — only the built-in skills.
- Scheduling is per-minute at finest. There's no sub-minute automation.
- Approvals expire after 24 hours (see above).
- It's experimental. If parts of the agent flow feel unfinished, that's expected at this stage.
The one thing to report immediately
If an agent ever takes an action outside the permissions you granted — say, a Read-only agent sends mail — that's a serious problem. Report it to the team with details. The permission scope is meant to be strictly enforced, always.
Related
How is this guide?
Last updated on