Voice calls
Starting a call with an agent, what the call screen shows, the four agent modes, and reading the transcript.
A voice call is a live conversation with an agent that produces a transcript as it happens. You can put the phone down and let it listen, or talk to it directly.
Starting a call
Tap Call on a record, or pick a record from Most Recent and choose a call. You are asked which agent, unless only one is attached to that record's entity type.
Three things happen before you are connected:
- If a voice note is recording, it is saved first. Recording and calling cannot run at once.
- The app asks for the microphone if it has not already. Refusing stops the call.
- The agent is started and the call connects.
Only one call runs at a time. Starting a second ends the first.
While you are connected
A line at the top reads "Connecting agent..." until the agent joins, then switches to "Connected" with a running timer.
Your phone treats it as a real call: it shows in the system call interface, keeps running when the screen locks or you switch apps, and can be ended from the lock screen. On Android an ongoing-call notification stays up for as long as the call lasts.
The call screen
Top to bottom: the agent's name, the mode pill and the participants; then the live transcript; then a floating bar with mute and end.
Leaving the screen does not end the call. A bar appears at the top of whatever you go to, and tapping it brings you back. Ending the call from anywhere closes the call screen and leaves you in the meeting's chat, where the transcript and everything the session produced is waiting.
Agent modes
The mode pill in the header switches how the agent behaves, mid-call, as often as you like.
| Mode | What the agent does |
|---|---|
| Listen | Nothing out loud. It records and builds context and stays silent. |
| Conversation | Answers when spoken to and never volunteers. This is the usual default. |
| Discovery | Facilitates actively, asking follow-up questions to draw out detail. |
| Recap | Synthesises, surfacing action items, decisions and next steps. |
A call opens in whichever mode the agent was configured to start in, and falls back to Conversation. A line in the transcript records every change, so you can see afterwards which parts of the call the agent was silent for.
Listen is the mode for a conversation the agent should not join, such as a patient consultation or a negotiation. Recap is worth switching to in the last minute of a call.
Reading the transcript
The transcript and the typed chat are one feed, in order. It can contain:
| Entry | What it is |
|---|---|
| Speech | What was said, attributed to whoever said it |
| A typed message | What you typed, with any photos attached |
| An agent reply | What the agent said or wrote |
| A report card | A report generated from this session, with its status |
| An artefact card | An extracted document, with its title |
| A video | A video you uploaded, with its processing status |
| A mode change | A line noting the agent switched mode |
Star a message to mark it. Lookups the agent runs appear inline in the transcript where they happened; reasoning is folded into a Thinking chip, which opens the full trace if you want it.
The transcript keeps itself current while you watch it, and catches up on its own if the connection drops for a moment.
Attachments
The composer takes photos and video. A video is uploaded and then analysed, and you can watch it move through uploaded, analysing and analysed on its card without leaving the transcript. The result lands in the Artefacts tab.
Everything else on a meeting
The header and its actions sheet also give you the participants list, a summary of the meeting so far, a prompt to generate a report, a way to regenerate the title from the content, and edit and delete for the session itself.