Working with apps

With your AMP organization linked and its apps connected (see Connected apps), just ask:

• "Find the e-mails from the auditors this month and summarize what they still need from us."
• "Post the release notes in the #product channel."
• "Update the stage of the three deals we discussed to Negotiation."
• "Save the attachment of the last invoice e-mail into this folder."

What happens

1. The agent looks up the app's actions and their inputs in AMP's catalog.
2. It reads what it needs. Reads run without asking.
3. It proposes the writes on an approval card: the app, the action and its exact inputs, up to 20 writes in one card. You approve, cancel, or answer in words.
4. After the writes run, the reply reports what each one did in its own terms ("sent", "updated"), checked against what the app actually returned.

Guard rails you'll notice

• Recipients who asked to be left out are blocked on the card.
• Many recipients (more than 10 by default) makes a write destructive: it asks in every mode.
• Outside domains: your organization can require a card, or refuse writes, to addresses outside its own domains.
• After reading outside data (an e-mail, a web page), every write in that turn goes to a card, and the card says what the turn read. See Untrusted data.
• Daily caps on writes can be set by your organization.

Files from apps

A file an app returns can be saved into your folder as a new file. A link from the app is downloaded straight from the provider to your Mac, over https to public addresses only.