A companion that goes first

Why this page exists

New in release 26.37

Live since September 10, 2026. The behaviour described here is switched on progressively. Reaching you off the app — by text message or call — only ever happens with your consent.

Most assistants wait for you to type. From release 26.37, Maggie can act before you ask — inside a boundary you set, and never without telling you. This page explains what "going first" means, what it will never do, and where the controls are.

The first act

When you start a conversation — especially your first one — Maggie has already done something useful with what is public about your business and what you told her. She says what she did, offers two or three things she could do next as taps, and says plainly what she would need in order to do more. You can ignore all of it and just type.

Two guarantees hold for that first act: it only ever uses public information and what you volunteered, and it is metered under a daily ceiling so a busy signup day cannot run up a bill.

Reaching you where you are

Magieva can reach you by text message or by phone, but only after you have said so.

• Verification is offered once, with a plain statement of what it enables. Decline and you are not asked again.
• The consent record is written at that moment, and every outbound message checks it first. If the record cannot be read — for any reason — nothing is sent.
• The first message carries opt-out instructions. Replying STOP works immediately.
• Quiet hours default to your timezone, and a nudge waits for your next free calendar slot rather than interrupting a meeting.

Watching on your behalf

You can hand Maggie a schedule ("every weekday at 8am…") or a condition ("tell me when…"). Routines and watches run without you, post into the conversation they were created in, and draw on a separate, lower spend ceiling than work you ask for directly. How to set them up and cap them is in Set up routines and watches.

Acting inside a boundary

When Maggie wants to do something you did not ask for, she first writes down a proposal: the action, its target, what it expects to change, whether it can be undone, what it will cost and why she thinks it is worth doing.

The proposal then meets your rules:

• If a standing rule you set covers it, it runs. Standing rules belong to the organization; they can be reviewed and revoked at any time, and every one appears in the activity feed.
• If no rule covers it, she asks — in the app, or by text if that is where you are. An unanswered proposal expires as refused.
• Anything irreversible always asks, whatever the rules say.
• Anything that moves external money is refused until an organization administrator has set a per-action ceiling, and refused above it.
• Unprompted work draws on the organization's unattended ceiling. When it is reached, unprompted work pauses; work you ask for directly is unaffected.

Two organization-level switches govern all of this, and unprompted execution needs both on. Turning either off returns Maggie to asking first for everything.

Every unprompted action can be reconstructed

For any action Maggie took on her own you can ask "why did you do this?" and get the chain: the signal that prompted it, the proposal, the rule or the approval that let it run, and what it cost. Members see the parts they have access to; nothing is hidden behind a blanket error.

What she will not do

• Send anything to you off the app without recorded consent.
• Act on instructions found inside content she was asked to read — an email, a web page, a document. That content is treated as untrusted.
• Spend past the unattended ceiling, or move money past the per-action ceiling.
• Carry out an irreversible action without asking.

The org's why, and the daily round

Live since 2026-09-15

Live in production since release 26.38 was promoted on 2026-09-15. The behaviour described in this section is switched on progressively.

Going first works better the more Maggie understands why your organization exists. From release 26.38, she drafts a short, plain-language statement of that — what you do, who it's for, how you work, what you value — starting from whatever is publicly visible about your business. Every line is marked unconfirmed until someone on your team agrees with it, asks her to try again, or corrects it directly. Nothing she infers is treated as true until a person says so.

Captured 2026-09-14 in a Magieva sandbox environment at /settings/charter, 1920×1080. Every statement in this specific screenshot happens to be at its starting (unconfirmed) state — a real organization will see these move to confirmed as people respond.

Once confirmed, this understanding grounds everything else Maggie does — including a short daily check-in, delivered into your regular conversation list rather than a separate inbox. It asks about one thing that matters that day, and your reply (or your silence) adjusts how often you hear from her about that kind of thing next.

Captured 2026-09-14, /settings/charter → "Your mornings" tab, 1920×1080. A fresh account with no replies yet — the "no weights yet" line is expected, not an error.

Both the org's why and the daily check-in are entirely optional to engage with: ignoring a check-in changes nothing except how often that kind of question comes back.

Where to go next

• Set up routines and watches, with spend limits
• Follow your organization's activity
• How Magieva works