Skip to navigation

Changelog

November 11, 2026

Email the caller

A new built-in tool. When someone asks “can you email me that?”, the expert sends them a follow-up email with the details they asked for, written from the conversation and the expert’s knowledge.

  • It is off until you turn it on for an expert, and it works for visitors too: share links, the website widget and phone calls.
  • On a call, the expert has the person spell their address, reads it back and sends only after a clear yes.
  • Every email has an unsubscribe link. An address that unsubscribes is never emailed by your workspace again.
  • Limits: one email per conversation, 3 a day per caller, 3 a day to any one address, 50 a day per expert and 200 a day per workspace.

See Tools.

Languages

Experts now answer in the language the caller actually speaks. They go by the caller’s words, not by the script the transcriber wrote them in, so accented English is answered in English. They follow the caller’s latest turn, back to English too, and keep names in their usual spelling. See Languages.

Knowledge

A document keeps up to 400,000 characters of text. A longer file is now marked Shortened on its row and in the library, with a warning when you upload it, so you know the expert doesn’t read the rest. Split the file and upload the parts to include all of it. See Library.

Experts

  • Paused experts say so. A paused expert’s public page now says it isn’t available, instead of Available now. See what pausing stops.
  • Back to draft is its own entry in History. Taking a live expert back to a draft is recorded as Back to draft, not as a publish. See Versions.

For developers

  • A new version reason, unpublish. GET /v1/agents/{id}/versions returns reason: "unpublish" when an expert was taken back to a draft. Versions from before 11 Nov say publish for that.
  • Keys limited to specific experts see less knowledge. Such a key can read only the shared knowledge folders and website imports that its experts use. See Scopes.
  • Chat grounded in a meeting needs meetings:read. Starting a conversation with meetingId from an API key now needs the meetings:read scope as well as chat:write.
  • Webhook test and resend return the delivery. Sending a test event and resending a delivery both answer { deliveryId, status, responseStatus, error }. See Webhooks.
  • Request body size limits. A JSON body can be up to 2 MB. Routes that carry a file or a whole document take up to 12 MB, and the public routes (/v1/public/…) take up to 1 MB. A larger body gets 413.
  • Chat messages are up to 32,000 characters. A longer content is refused with 400. Visitors’ messages stay at 4,000.