Say what that mail should become
Create a workspace named for the work, such as one kind of customer message or one kind of notice. In a blank workspace, tell AI which mail to recognize, what the render view should keep, and which cases must wait for you. If a template is close, load it and then change those three points.
The difference from a one-off draft is that the result stays in the interface, data, and files, and the next session continues from there.
Connect a mailbox
The app has a mail category. Open the connection screen the build actually shows, choose the mailbox provider, and complete the sign-in there. Do not assume a server, port, or permission scope before you see that screen.
Before the connection exists, you can still build the categories, summaries, and open items from sample messages you place in the workspace, and check them in the render view.
Let AI follow one set of rules
When the sample structure works, ask AI to handle new messages in that structure: write them into the workspace, mark items that need your confirmation, and keep the original points. When the same pass should repeat, create a scheduled task, choose an agent session, and point the prompt at those rules.
Use run now on one result first. If it is wrong, change the rules in the workspace rather than correcting them only inside one chat.
FAQ
Can I start before a mailbox is connected?
Yes. Build the viewing and handling structure from samples first. Use the app's current flow for the live connection.
Can AI send mail on its own?
The mail tools can draft and send. Write the rule that sending waits for your confirmation, keep that rule until a few runs are correct, and only then decide whether to loosen it.
How does this relate to a knowledge base?
Conclusions that should last can be organized into a knowledge base. The email workflow handles this class of mail.
Next step
Create a workspace from a real sample of one kind of mail, make the open items readable in the render view, then connect the mailbox with the flow the app shows now.
Internal links
- AI email triage workflow
- Extract invoices from email
- How to schedule recurring AI tasks
- How to do AI knowledge management
- Browse workspace templates