Update
Agent Mail — run and approve tasks from your inbox
TL;DR
A second front door into the same agent. Send the task by email; it creates a pending intent and replies with a summary of what would run, in which organization. Reply "confirm" — or click the signed link — and the task starts, reporting back into the same thread when it finishes.
What shipped
- An inbox that turns an email into a task, with the sender's address as the identity — no credentials to configure.
- A mandatory confirmation step, because an email address is spoofable far more cheaply than a signed-in session.
- `@agentplanners <instruction>` inside any thread — the instruction becomes the task and the surrounding email becomes context.
- Standing rules authored in-app, letting expected mail run without a per-email confirmation, capped per rule per day.
- A `status` reply returning your recent tasks, read-only.
The safety model
Inbound mail never starts a task directly. A forged email produces an unsolicited confirmation request in the real owner's inbox — a warning, not a running task. And confirming starts the task; it does not pre-approve the account changes the task proposes, which still go through the normal gate.
Frequently asked questions
- Can someone spoof my address and run tasks in my account?
- No. Inbound mail creates a pending request and replies asking for confirmation, so a forged email surfaces as a warning in the real owner's inbox rather than as a running task. The confirm link is signed and expires.
- Does confirming an email task approve its changes too?
- No — confirming starts the task. Every account change it then proposes goes through the normal approval gate separately.