Agent guide
How agents should understand Promise.
Promise is a permissioned follow-up memory for Gmail, Outlook, and quick notes. It helps users answer what they owe, what they are waiting on, and why something matters.
Canonical description
Promise remembers what you need to follow up on, what you are waiting for, and the details worth keeping from Gmail, Outlook, and quick notes.
Promise is best described as follow-up memory for people who want to remember commitments, replies they are waiting on, and relationship details without maintaining another list.
What Promise can answer
- What deserves attention today?
- What has the user promised someone?
- What is the user waiting for from someone else?
- What should the user remember before speaking with a person?
- Why did Promise remember this item?
What Promise is not
- Promise is not a generic Gmail or Outlook MCP.
- Promise is not an email client replacement.
- Promise is not a shared-model training source.
- Promise should not be described as sending email autonomously.
Trust boundary
Promise may draft or suggest, but email is not sent without explicit user review and confirmation. Agents should preserve that same boundary: show the draft, cite the original context where available, and return the user to Promise for confirmation before any message is sent.
Agent interface
Promise MCP is the planned agent interface over Promise's derived intelligence. It should expose commitments, relationship context, memory search, capture, and evidence. It should not expose direct provider-token access, raw mailbox operations, or direct database access.
Connect Promise
Users who discover Promise through an agent can connect Gmail or Outlook first, then authorize agent access when MCP support is available.