Update
Agent Mail gets HTML replies, CSV attachments and per-organization ingest
TL;DR
The email channel's first release proved the flow; this one makes what lands in your inbox actually usable. Reports come back as formatted HTML rather than plain text, the underlying rows arrive as a CSV attachment you can open in a spreadsheet, and mail ingest is scoped per organization.
What shipped
- HTML replies — a report emailed back into the thread renders as formatted output rather than a wall of plain text, with a plain-text alternative retained.
- CSV attachments — the data behind a report arrives as a file, so the numbers are usable without copying them out of an email body.
- Per-organization ingest, so mail routes to the right workspace rather than relying on a single shared path.
- A bounded reply to unknown senders — the pointer sent to an address that isn't a member is rate-limited, since an unknown sender has no organization to bill and shouldn't be able to generate unlimited replies.
- Forwarded-mail bodies handled, and mention/clarification text fenced so quoted content can't be misread as an instruction.
Why the fencing matters
An email body is untrusted input written by whoever sent it. Fencing the mention and clarification text keeps quoted or forwarded content as data the agent reasons about rather than as instructions it might follow — the same discipline described in is it safe to let an AI agent read your email.
Frequently asked questions
- Can I get the data behind an emailed report?
- Yes — the rows arrive as a CSV attachment alongside the formatted HTML reply, so you can open them in a spreadsheet without copying anything out of the email body.
- What happens if a stranger emails the agent inbox?
- They get one polite pointer explaining the address isn't linked to an account, and that reply is rate-bounded — an unknown sender has no organization to bill, so unlimited replies would be our cost to absorb.