AI follow-through should come with receipts.

The problem is not that people forget to make lists

Commitments usually disappear because they are made in normal conversation. A client asks for a follow-up next week. A teammate says they will send an update. You remember a personal detail while walking between meetings. None of that starts as a task-manager entry, and most people do not want another list to maintain.

Promise starts from that reality. Gmail, Outlook, and voice notes are capture mechanisms. The product job is not to replace your inbox or calendar; it is to help you follow through on the commitments, requests, dates, and details already flowing through your day.

Generated tasks create false confidence when evidence is missing

AI can turn a conversation into a plausible action item. That is useful, but it is also risky. If the extracted task is detached from the source message, voice note, speaker, date, and surrounding context, the user has to trust a summary without being able to verify it quickly.

Promise is designed around memory with receipts: when it surfaces a follow-up, promise, or remembered detail, it preserves the original evidence where possible so the user can inspect why the item exists before acting on it.

Good follow-through needs attribution, timing, and correction

A useful reminder is not just text. It needs to answer: who is this about, what was committed to, where did it come from, why now, and is the system wrong? Promise prioritizes those boundaries because relationship memory is intimate and mistakes should be easy to spot.

That is why Promise treats correction as part of the product, not an edge case. Users should be able to review evidence, dismiss what is not useful, correct remembered details, and keep the underlying mail app as the place where final communication happens.

Voice and email are inputs, not competing products

Voice capture fits the same model as email: say the small thing while it is fresh, then let Promise bring it back when it matters. A note like "follow up with Kevin in a week" should become a timed follow-up, not another unread note buried in a recorder app.

Email works the same way. Promise looks for commitments, follow-ups, dates, and relationship details inside conversations you already have, then shows them in a calm Today view with source context.

Explicit confirmation is the trust boundary

Promise may help draft a response, but it does not send generated email without explicit user review and confirmation. That boundary is central to the product: AI can help recover context and propose the next step, while the user remains responsible for the final action.

For the same reason, Promise explains what inbox access means before authorization and keeps privacy, security, disconnect, export, and deletion details available from public pages.

What Promise does today

Promise currently helps identify follow-ups, open commitments, reply opportunities, dates, and important details from Gmail, Outlook, and voice notes. It brings those items into Today, links them to evidence where possible, and lets users review the source before responding or marking work complete.

Learn more in How Promise works, Security, and Privacy.

What Promise does not do

Promise is not a replacement email client, not a generic chatbot, and not an autonomous sender. It should not invent obligations without review, claim certainty when evidence is weak, or ask users to trust a generated task that cannot be traced back to its source.