Messages arrive
Mail, chat, Apple Messages, and work updates flow into the inbox.
Agents
Think of them as careful helpers for the message pile.
An agent is not a robot taking over your life. It is a supervised helper with a job: read the incoming mess, notice what matters, prepare the next step, and ask you before doing anything consequential.
Drafts replies
Finds next steps
Asks before acting
Remembers context
Agents read the pile
They look for urgency, people, promises, dates, and useful context.
They prepare work
Summaries, drafts, tasks, searches, and suggested rules are created.
You decide
Policy gates keep sends, deletes, and recurring changes in your review flow.
What it can do for me
Useful jobs an agent can take off your plate.
Each agent has a role, like a careful assistant who knows the difference between preparing work and making decisions for you.
Morning sorter
Shows you what actually needs attention before the day starts running you.
Tech note: Triage means the agent ranks messages by urgency, sender, deadlines, and past behavior.
Reply drafter
Writes the boring first version so you can edit instead of starting from a blank box.
Tech note: Draft-only mode means the agent can prepare text, but sending waits for your approval.
Follow-up keeper
Turns loose promises into reminders, tasks, and calendar-aware next steps.
Tech note: Structured extraction means dates, people, and action items are pulled from plain messages.
Junk reviewer
Keeps newsletters and repeat clutter from crowding out the people you care about.
Tech note: Feedback learning means your corrections become preferences the agent can use next time.
Memory helper
Remembers who someone is, what was promised, and where the last conversation left off.
Tech note: A private knowledge graph links people, companies, topics, and source messages.
Tool runner
Looks things up in approved tools so replies are based on real information, not guesswork.
Tech note: MCP connectors are supervised tool bridges: agents can use them only within the limits you set.
You can see who is working on what.
The team dashboard keeps agents visible: online status, recent reports, pending proposals, and where they are blocked.
Tool use is shown, not hidden.
When an agent needs a tool or approval, the request appears in the conversation so you can understand the action before saying yes.
Next step
Meet agent employees
When an agent has a real job, it becomes an employee.
An employee has a name, a role, a mailbox or channel, allowed tools, and a visible activity log. You are the boss: you set the job, review proposals, and decide how much independence each employee gets.
The safety bits
A little technospeak, translated.
The magic is less magical when you can see the safety rails.
The agent can prepare, but you approve high-stakes actions.
A support agent and a household admin agent can have different permissions.
Good answers point back to the messages or tools they came from.
Simple jobs can use cheaper or local models; harder writing can use stronger ones.
Start with messages. Let agents prepare the work.
The inbox feeds the agents; the agents feed you a smaller set of drafts, reminders, summaries, and decisions.