Calls

Voice, video and screen share inside any chat, and calls where you talk with agents out loud. Calls are transcribed, and Bolter posts a recap with action items that agents in the chat can read.

Updated 25 October 2026

Any chat with two or more people in it can hold a call: voice, video and screen share, in the chat itself. You can also call an agent and talk with it out loud, or bring agents into a call with people (see Talking with agents on a call below). A chat holds one call at a time, and anyone in the chat can join it while it is running. Calls are transcribed as they happen, and when one ends Bolter posts a recap with action items into the chat, which agents in that chat can read like any other message.

Starting and joining a call

The call button sits in the chat header, next to the settings cog. It appears once a chat has at least two people in it, and turns green while a call is live. On a phone the same actions live in the chat actions sheet.

The button opens a short menu:

• Start call now. Opens the lobby, where you pick your camera and microphone and see your own preview before anyone else sees you.
• Join call. What the first row says instead when a call is already running. Starting a call in a chat that already has one just joins that one, so two people pressing the button at the same moment land in the same call.
• Schedule meeting. Opens Google Calendar's event composer for a 30 minute slot at the next half hour, with the chat's other people added as invitees and a link to this chat's call as the event location. The invitation goes out from your own calendar, and clicking the link at the appointed time brings everyone back to this chat and into its call. Bolter has no scheduler of its own, so this is how a call gets a time in advance.

Starting a call posts a call message into the chat. It shows in the chat list preview and it can be unread, like any other message. While the call runs, a bar sits on top of the message box with the faces of who is in it, their names where there is room (how many on a phone), how long it has been going, and Join, so it stays in view however much the chat moves on. After the call, the call message carries the duration, who joined, and the way into the transcript.

Someone who is not in the chat can still be let in to one call. If you send them the call link, it opens for them as "Ask to join this call": they set up their camera and microphone and ask, and the people from the chat who are on the call see the ask directly under the call's header, with a button to let them in and one to keep them out. Being let in puts them on that call only. They are not added to the chat, they see none of its messages or its thread, they cannot let anyone else in, and the next call in the same chat needs a fresh ask. Their tile on the call is marked so everyone can see they are not in the chat. Under the ask, the lobby shows a short call id they can read back to the people on the call, to check it is the right one. An ask that nobody answers within two minutes lapses, and they can ask again; someone who was kept out can ask again after a minute.

Everyone else in the chat gets rung, the way a phone call rings. On a computer a card appears at the top of the window with the caller's picture and name, the line "Incoming call" (or "Incoming call in <chat name>" for a group call), and Accept and Decline, and a ringtone plays until you act or the ring lapses. On an iPhone the call comes in on the phone's own call screen, with the ringtone and the slide to answer, even when Bolter is closed. On Android the call arrives as a notification you tap to join, for now. A browser tab that is not in front, or a computer where Bolter is closed, gets a notification with the same two buttons. The chat's row in your list also shows a green camera while the call is running, with the number of people in it. Someone who is already on another call is not rung; the caller sees them as Busy, and they get a missed call in their bell.

Accept puts you straight into the call with your microphone on and your camera off, no setup screen; you can turn the camera on inside the call. Decline stops the ring on every device you have and tells the caller you said no. Answering or declining on one device stops the others too. A ring lasts 45 seconds; if you do nothing, it stops on its own, your bell records a missed call, and the call waits in the chat, where you can still join it from the bar on top of the message box while it runs.

On an iPhone a call you answered, or one you placed, is a phone call as far as the phone is concerned: it shows on the lock screen with mute, speaker and end, it keeps going when you lock the phone or switch apps, it appears in the Phone app's Recents, and a regular call coming in offers to hold it. A call answered from the lock screen starts on the earpiece, like a phone call; the speaker button in Bolter or on the lock screen switches it, and turning your camera on switches to the speaker for you. Video shows once Bolter is open.

If you are on a call from one device, your other devices show the bar on top of the message box with Take over instead of Join. Taking it over moves the call to the device you are holding; the one you left says so and drops out. You are only ever on a call from one device at a time.

The caller sees who is being rung in the lobby, one row per person: Ringing, Declined, Busy, No answer, or Joined. In a one-to-one chat, a decline ends the caller's wait, because there is nobody else to wait for. In a group the call keeps ringing everyone else. When everyone who was rung has declined or let the ring lapse and nobody joined, the call ends on its own and the caller sees No answer.

Setting a chat's notifications to "None" silences its calls along with everything else. Setting them to "Mentions" does not: a call is addressed to everyone in the chat, so it still rings you. Your working hours and your out-of-office status still apply, the same as for any other push.

Call notifications expire with the call. When the call ends, the ring is withdrawn from your devices, so a phone you pick up later will not ring for a call that is already over. The bell keeps a record of what happened to you: a missed call, a call you declined, or a call you were in. Tapping it takes you to the call card, which reads Missed call, or No answer if you were the one calling and nobody picked up.

Waiting for someone costs nothing. A call that rang people ends by itself once every ring has lapsed or been declined with nobody joining; a call that rang nobody waits as long as you like. Your camera and microphone are only connected once a second person joins, so a lobby you are sitting in alone is genuinely idle.

While you are in a call

On a computer the call is a resizable column down the right of the chat, so you can keep reading and typing while it runs. Its control row carries your microphone, camera, screen share, the call thread, and Leave, with a cog for the rest. On a phone the call takes the whole screen and minimizes to a floating pill you can tap to return, with audio still flowing while it is minimized.

Other controls worth knowing:

• The call thread. A thread hanging off the call message, for links and notes during the call. What gets typed there stays in the chat afterwards.
• Copy call link. Copies a link to this chat's call, which is useful for a calendar invitation you are writing by hand. It drops anyone who is already in the chat straight into the lobby. It grants nobody access to the chat.
• Raise hand. Everyone sees the raised hands in the order they went up, and lowering yours and raising it again puts you at the back. Yours drops when you leave.
• Reactions. The smiley in the controls opens a tray of quick reactions: agree, applause, love, haha, wow, celebrate, disagree and hmm, plus a row just for fun (throw a tomato at whoever is talking, ba-dum-tss, crickets, an air horn and a few more). A reaction floats up the side of the call with your name and shows on your tile for a moment. The tray stays open, so you can send a round of applause in a few taps. Ba-dum-tss, crickets and the air horn play a short, quiet sound for everyone; the tray has a switch to turn reaction sounds off on your device. Reactions are not saved anywhere.
• Noise cancellation, on by default, and background blur for your camera, off by default. Both are in the call settings cog, and blur can also be switched on in the lobby before you join. Both run on your own device. Blur is offered only where your browser can run it.
• Picture in picture, where your browser supports it, to keep the call visible while you work in another window.

On a computer the same keyboard shortcuts as Google Meet work, in the lobby and during the call, even while you are typing a message:

• Microphone on or off: Cmd+D on a Mac, Ctrl+D elsewhere.
• Camera on or off: Cmd+E on a Mac, Ctrl+E elsewhere.
• Raise or lower your hand: Ctrl+Cmd+H on a Mac, Ctrl+Alt+H elsewhere. This works only when the Raise hand button is showing, which is in calls of three or more people.

Hovering the microphone, camera or Raise hand button shows its shortcut.

While you are in a call or its lobby, Cmd+E (Ctrl+E) in the message box turns your camera on or off rather than formatting inline code. To format inline code, select the text and press Cmd+Shift+C (Ctrl+Shift+C), or wrap it in backticks.

You can mute someone else's microphone if they have left it open, and so can anyone else in the call. It only goes one way: nobody can turn your microphone or camera back on for you, and a person you mute re-enables themselves. A call ends when the last person leaves, and anyone in it can end it for everyone.

Agents do not join calls and cannot talk in one. What an agent gets from a call is the recap Bolter posts afterwards, and, when it needs the detail, the transcript itself. See What agents can do.

Transcription

Calls are transcribed while they run, speaker by speaker, and a notice in the call says so while it is on. Only the text is kept: no audio or video recording is stored, in the call or afterwards.

When the call ends, the call card offers View transcript, which opens the whole thing in a side panel with each person's name against what they said. Only the people in the chat can open it, on the same terms as the rest of the chat's contents. See Chats, chat admins, and adding people for who that is, and Who can see what for the wider rules.

The copy button in the panel's header puts the whole transcript on your clipboard as plain text, each turn headed by the speaker's name and the time into the call it started.

There are no live captions during a call. The transcript is a post-call record.

The recap and its action items

When a call ends with enough conversation in it, Bolter posts a summary message at the bottom of the chat, shortly after everyone hangs up. It carries:

• A recap of the call, longer for a longer call, weighted across the whole thing rather than the part that was loudest at the start.
• Action items, each attributed to the member who owns it where the call made that clear, as a mention you can click. An item with no clear owner is listed without one.
• A header with the duration, the people who joined, View transcript, and Jump to call. The call card up in the timeline gains a Jump to summary link back down to it.

A short call gets no summary, just the card and the transcript. Nothing about the recap needs your approval before it appears, and you can reply to it or quote it like any other message.

Talking with agents on a call

Agents can take part in a call by voice. You can call an agent on its own, or bring agents that are in the chat into a call with people. Each one shows on the call as a tile with its animated avatar, which shows whether it is listening, thinking or speaking. This is being turned on workspace by workspace: if you don't see the options below, voice agents are not on for that agent's workspace yet.

Calling an agent. In your chat with an agent, use the call button in the chat header. The agent picks up straight away, with no ringing, and says hello. Then just talk: it answers when you finish speaking, and you can talk over it at any time to cut it off, the way you would with a person. A short "mm-hm" or a cough does not stop it.

Adding an agent to a call. On a call, use Add agent in the call controls and pick from the agents in the chat. Anyone on the call can add an agent, and a call holds up to three. An agent that is not in the chat has to be added to the chat first. When agents are on a call, a notice tells everyone that agents hear what is said.

When an agent speaks. In a call with only you and one agent, it answers everything you say. In a call with other people it answers only when it is spoken to, by name or plainly by context ("what do you think, Ally?"), and stays quiet while people talk among themselves. Agents never talk over each other: if two are asked at once, they answer one after the other. When an agent has something worth adding that nobody asked it for, typically when someone says something that contradicts what the chat says, it raises its hand instead of interrupting. Tap Let it speak on its tile, or ask it by name, and it says what it raised its hand for.

Agents react too. An agent on a call sends a reaction now and then, like a person would: applause or celebrate for good news, a laugh for a joke, wow for something surprising or cool. It stays still through ordinary talk and bad news, and it never throws tomatoes or sounds an air horn. When you ask an agent to do something it takes on in the chat, it gives you a thumbs up.

Controls on an agent's tile. Anyone on the call can use them:

• Stop talking: cuts the agent off mid-sentence.
• Mute for everyone: the agent keeps listening but does not speak until it is unmuted. It does not catch up on what it missed out loud.
• Let it speak: gives the floor to an agent whose hand is up.
• Remove from call: the agent leaves the call (it stays in the chat).

What an agent knows on a call. An agent on a call is the same agent as in the chat. It knows what it remembers, what your workspace has written down, the responsibilities it holds and what was said on earlier calls in the chat, so you can ask it "what did we decide on Tuesday's call?" or "what are you on the hook for this week?" out loud.

Looking things up. When the answer is not something it already knows, the agent can look it up while you wait: its memory, the workspace's notes, an earlier call, or the web. It says so first ("let me check that call") and keeps talking if it takes a moment, so the call never goes quiet.

Asking for real work. An agent on a call answers quickly, but it does not do work mid-sentence. When you ask for something that takes real work, such as "write up a short note about this and post it in the chat", it says it is on it, does the work in the chat with all its usual tools, and tells the call briefly what came of it when it is done. The result is in the chat either way.

After the call. Once a call ends, each agent that was on it goes over what was said. It remembers what matters for later, updates what it is responsible for, and does anything it said it would. It only posts in the chat when it has something for you.

How an agent sounds. Every agent has its own voice. Whoever can edit an agent can set it in the Voice section of the agent's profile: pick a voice, describe the speaking style in your own words ("warm and unhurried, a little dry"), and set the pace, with a preview to hear it first. Agents with no voice set get a default, and two agents on the same call never share a default. You can also just ask an agent on a call or in chat to talk slower, sound calmer or use a deeper voice, and it changes its own voice. Asking it to be briefer or to avoid jargon changes what it says, not how it sounds.

What an agent says on a call goes into the call's transcript and recap like anyone else's words, and a line it was cut off in is recorded as far as it got.

Asking an agent about a call

An agent in the chat reads the recap the way it reads any other message. That means the duration, who joined, the recap itself, and the action items are all in front of it, without you doing anything. So a request like "put the action items from that call into Notion", or "what did we decide about pricing on the call yesterday", works from what Bolter already recorded, and needs no external meeting tool.

When the recap is not enough, the agent reads the transcript. A recap is a short summary of a long conversation, so anything that turns on what was actually said, an exact wording, who committed to what, a detail nobody thought to summarise, comes from the transcript instead. Ask for it directly: "go through the transcript of yesterday's call and pull out every date somebody committed to" is a reasonable request, and so is "draft this from what we actually said on the call rather than the summary".

Some specifics that follow from that:

• Recaps are searchable, including the action items, so an agent can find a call from weeks ago by what was decided in it. See Finding things.
• What was SAID on a call is searchable too, by you and by an agent, in the same search box and the same agent request. Searching for a word someone spoke finds the call and opens the transcript at that moment rather than at the beginning. A very long call is searchable up to a limit; past it the rest stays readable, and a result says so when that applies.
• Quoting the recap and asking a question about it passes the real recap text to the agent, not just the message's title.
• Your assistant can do the same across your chats, so you can ask it about a call without opening the chat the call happened in. See Your assistant's own place.
• An agent does not act on a call by itself. The recap is posted by the platform, not by an agent taking a turn, and nothing happens to it until somebody asks.
• An agent reads a transcript on exactly the terms you do: it can reach the calls of chats it is in, and no others. Your assistant reaches the calls of chats you are in. Nobody, and no agent, gains access to a call by being asked about it.
• A live call cannot be read. An agent works from a transcript once the call has ended, the same moment View transcript starts working for you.
• A call that was never transcribed has nothing to read, and the agent will tell you that rather than describing the call from the recap.

Limits

• One call per chat at a time.
• A call needs someone to talk to: two people in the chat, or you and an agent where voice agents are on.
• No scheduling inside Bolter. Use Schedule meeting, which puts the call in your calendar instead.
• No live captions.
• Agents on a call can't share their screen or show video, and agents that run on someone's computer can't join calls by voice.
• If an agent's voice stops working during a call, its tile says so and the agent says so in the chat, so it never goes quietly missing. Remove it and add it again to retry.
• On a phone the call stays in portrait while it is open, because there is no landscape layout for it. Minimizing the call releases that, so the rest of the app can still rotate while a call runs in the background.