View AI answers in Flow chat with show_in_chat

Ask the AI questions about your meeting — while it's happening or after — and get answers in the chat instead of scribbled over your notes.

How it works in Flow

Overview

The AI reads your meeting transcript and responds in three ways: answering questions in chat, rewriting your notes, or making targeted edits to your notes.

Key behaviors

• Questions stay in chat: Asking "what did he just say?" or "can you summarize the last few minutes?" returns the answer as a chat bubble and never touches your notes. The Thoughts tab changes only on an explicit instruction, such as "update my notes with the action items just discussed."

• Works in both views: Chat from the floating recording window during a live meeting, or from the full Notetaker view.

• Live and post-meeting context: The AI uses the live transcript during the meeting and the refined transcript, with speaker names, after it ends. It also knows the current date, time, and timezone, so "what's today's date?" and time-bounded questions about your meetings and notes resolve correctly.

• Grounded answers: Responses lead with the answer, citing what was specifically said or decided, and state plainly when something wasn't covered in the transcript. Answers are typically under 250 words, and lists display with bullet points or numbered markers.

• Reliable for large or cross-meeting questions: When a question spans multiple long meetings, the AI processes each transcript and combines the results, so no content is dropped and meeting citations still work. A meeting whose details can't be read is noted as unavailable rather than silently skipped.

• Citation pills: Sources cited in an answer appear as pills below the reply, each showing the meeting's participant photo, a note icon for standalone notes, or a colored initial if no photo is available. Meeting titles display in full and wrap rather than being cut off.

• Answers appear word by word: New replies reveal word by word with a blinking caret, preceded by an animated loader. Chats already stored on your device — including when switching chats in the Hub or the meeting chat switcher — appear instantly; only chats that exist solely on the server show a brief loader. You can load earlier turns in a conversation at any time, including when reopening a chat.

• "What did I miss?" chip: While a meeting is recording, the What did I miss? chip in the chat composer returns a recap of the last ~2 minutes as a chat bubble, or "Nothing new since you last looked." if nothing meaningful was said. The chip is disabled while another reply is loading. Clicking it expands the chat drawer to a readable height if it was too small, but never shrinks a drawer you resized larger.

• Chat drawer opens and closes: The drawer in the notepad footer opens from the chat input toggle and closes when you click outside it, such as on your notes or the transcript. Clicking inside the drawer or on the toggle keeps it open, and it returns to your saved height when reopened.

• Scroll stays pinned while streaming: The chat stays scrolled to the bottom as a reply streams in. Scroll up to read earlier messages without being pulled back down; sending a new message snaps the view to the bottom.

• Plain error messages: If the AI can't reach the servers, the chat shows "Sorry, I'm having trouble reaching the servers right now — please try again in a moment." instead of a red error. Other failures show a message matching the failure type, such as "Could not get a response. Try again." or "Something went wrong. Try again."

Examples

FAQs

Limitations and notes

• Notetaker is in beta, available on Mac only, and rolling out gradually — it is not on Windows or mobile.

• The chat AI answers questions in chat, rewrites your notes, or makes targeted edits to your notes — it does not create standalone documents.

• Answers are based on the meeting transcript — the more complete the transcript, the more accurate the answers.

• The AI knows who attended a meeting based on your calendar, but speaker attribution in answers comes from the transcript itself, not the attendee list.

• Complex questions spanning multiple long meetings take noticeably longer to answer; the AI completes the response rather than failing silently.

• The "What did I miss?" chip is only available during a live recording, not after a meeting has ended.

• Only one chat request runs at a time — the chip and the composer are disabled while a reply is loading.

• The auto-expand from "What did I miss?" is temporary — your saved drawer size is restored when you reopen the notepad or restart the app.