Made in 10Docs
GuidesAPI
Calls

Meetings

Send a character into a Zoom, Google Meet or Microsoft Teams meeting by its link, now or at a time.

A character can join a meeting the way a person does: give it the meeting's link, and a meeting bot joins with the character's face as its camera and its voice as its sound. The meeting becomes a call with source: "meeting", charged, analysed and remembered like any other.

You can send one from the dashboard (Meetings, then Send to a meeting: pick the character, paste the link, choose now or a time in your own time zone, and set the bot's name, how long it may stay, when it speaks, recording and who it remembers. The page lists live, upcoming and past meetings; open one to cancel it, move or rename it while it waits, or read its conversation) or through /v1, below. The dashboard's meetings are live ones.

Send a character into a meeting

POST/v1/meetings

Prop

Type

It takes an Idempotency-Key header, as starting a call does: the same key gives back the same meeting (with 200 instead of 201). See Idempotency.

curl https://api.daxov.com/v1/meetings \
  -H "Authorization: Bearer $ACME_SECRET_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: 8d2e4c1a-6b7f-4e90-a3d5-0f1c2b3a4d5e" \
  -d '{
    "character_id": "chr_4fT9kQ2mXw7LbN3pZs8V",
    "meeting_url": "https://meet.google.com/abc-defg-hij",
    "join_at": "2026-11-06T15:00:00Z"
  }'

meeting_url must be an https link to a meeting on:

  • Zoom: *.zoom.us or *.zoomgov.com, with a path under /j/, /w/, /my/ or /wc/;
  • Google Meet: meet.google.com/abc-defg-hij;
  • Microsoft Teams: teams.microsoft.com, teams.live.com or teams.cloud.microsoft, with a path under /l/meetup-join/ or /meet/.

No user name in it, no other port, never an IP address, at most 2,000 characters. A #fragment is dropped. Anything else is 400 invalid_meeting_url.

The time

join_at more than 5 minutes in the past, or more than 90 days ahead, is 400 invalid_field.

One character, one meeting at a time

A character can't be sent twice into the same meeting at the same time. A second meeting for the same character and the same link whose time overlaps an open one (counting each one's max_seconds and 10 minutes in the waiting room) is 409 meeting_exists, and error.meeting_id is the meeting already going:

409
{
  "error": {
    "code": "meeting_exists",
    "message": "this character is already going to that meeting at that time; change or cancel that one",
    "meeting_id": "mtg_8Rk2Vn5XqL9mTb3PcW7z"
  }
}

The same link on another day, another character in the same meeting, or a test meeting beside a live one are each their own meeting. To send the same meeting again safely after a timeout, use an Idempotency-Key.

The name

Every bot's name ends with "(AI)". A bot_name that doesn't gets " (AI)" added (Notes becomes Notes (AI)); without a bot_name, it is the character's name followed by " (AI)".

The meeting

Meeting
{
  "id": "mtg_8Rk2Vn5XqL9mTb3PcW7z",
  "character_id": "chr_4fT9kQ2mXw7LbN3pZs8V",
  "character_name": "Sara",
  "meeting_url": "https://meet.google.com/abc-defg-hij",
  "platform": "google_meet",
  "bot_name": "Sara (AI)",
  "status": "scheduled",
  "reason": null,
  "error": null,
  "waiting_room": false,
  "waiting_since_unix": null,
  "join_at_unix": 1793977200,
  "max_seconds": 3600,
  "record": true,
  "respond": null,
  "people": null,
  "call_id": null,
  "env": "live",
  "metadata": {},
  "memory_key": null,
  "created_unix": 1793900000,
  "joined_unix": null,
  "ended_unix": null
}
  • platform: zoom, google_meet or microsoft_teams.
  • status: scheduled, joining, in_meeting, ended or failed.
  • reason: why it ended or failed (below); null until then.
  • error: a plain sentence for a failed meeting, such as "Nobody let the AI in from the waiting room within 10 minutes. Nothing was charged.", and for one cut short (balance_limit, interrupted). Null otherwise.
  • waiting_room: true while the bot waits to be let in. waiting_since_unix: when it first waited.
  • max_seconds: what you asked for until it starts; then what it may stay, after the balance or a test key capped it.
  • record and respond: as sent, or their defaults.
  • people: how many other people the bot last saw in the meeting (not counting itself or other meeting bots); null until it knows.
  • call_id: the meeting's call, once it has one. Get the call for its transcript, charge, analysis and recording.

What happens

Made

The link, the time and the character are checked, and whether a call could start at all: 409 price_not_set or tier_unavailable, 403 org_disabled, and 409 meeting_exists for a second meeting at the same time. The meeting is scheduled. Nothing is held yet, and it doesn't count against your calls at once.

Sent on its way

Without join_at, at once. With one, 15 minutes before for a voice-only character, and 2 minutes before for a character with a face. It becomes a call with source: "meeting": the same checks and hold as POST /v1/calls, and the same call.started webhook. The balance is held then, not when the meeting is made: a meeting the balance doesn't cover at that moment fails with insufficient_balance, and one it covers only in part is capped to what it covers.

A meeting due now is sent before the answer, so its refusals are the answer and nothing is kept: 402 insufficient_balance, 429 too_many_calls, 422 meeting_not_found (the bot couldn't use the link), 503 engine_unavailable or call_not_started.

A scheduled meeting that can't go when it's due is failed, with the reason. One that can't go yet (your calls at once are full, or no bot is free) is tried every 30 seconds, until 10 minutes past its time.

Joining

The bot is on its way in. waiting_room is true while it waits to be let in, for 10 minutes at most, and meeting.waiting_room is sent the first time it waits.

In the meeting

The bot is in, and meeting.joined is sent. The call is charged per second from when the AI joined, exactly as a call is: the waiting room is free.

Over

ended when it was in: everyone left, the host ended the meeting or removed the bot, its time ran out, the balance ran out, or you called it off. failed when it never got in. meeting.ended or meeting.failed is sent, and the call settles on its own with call.ended.

In the meeting

It says it's an AI, and that it's recorded. Its first line says its name and that it's an AI assistant. When the meeting's call is recorded (record), that line also says the meeting's audio is being recorded, and once the bot is in, a message in the meeting's chat says so too: "Sara (AI) is an AI assistant. This meeting's audio is being recorded." See AI disclosure.

It waits for a pause to greet. It doesn't talk over people already speaking when it joins: it greets at the first pause, or after a minute at the latest. If someone speaks to it first, its answer starts with that introduction.

In a group, it speaks when spoken to. With respond: "when_addressed", or with respond null once more than one other person is in the meeting, it answers only:

  • when someone says its name (any word of the character's name, or the first word of the bot's name);
  • when someone says "AI" or "assistant";
  • when someone talks over it, or speaks within a few seconds after it finished.

Everything else said in the meeting is in the call's transcript, and the character reads it before its next turn, but doesn't answer it. With respond: "always", or one other person, it answers everything, as on a call. Characters on a speech-to-speech tier take their own turns: they answer everyone and greet at once.

It leaves when it's alone. Silence is normal in a meeting, so silence never ends it. It leaves when no other person has been in the meeting for 60 seconds: everyone left, or only note-takers and other meeting bots remain. It also leaves at max_seconds, or when the balance it held runs out.

It can't see. It hears everyone as one voice; it doesn't see the people, the screen or the chat.

Recordings. The company that runs our meeting bots keeps no recording of its own: no audio, video or transcript is stored with it. Only we record, and only when record is on. See Recordings.

If we lose it. Should our side restart during a meeting, its bot is taken out of the meeting. A meeting that hadn't got in yet is sent again with a new bot (on a new call), while it's within 10 minutes of its time. One that was in ends interrupted, and nothing is charged for its call.

Reasons

reasonStatusWhy
completedendedIt was in and the meeting ended, everyone left, or only other bots remained
time_limitendedIt stayed its max_seconds (or a test key's 60)
balance_limitendedThe balance set how long it could stay, and that time ran out
removedendedThe host removed it
cancelledendedYou called it off
interruptedendedWe lost it partway through. Nothing was charged
not_admittedfailedNobody let it in within 10 minutes, or it was removed from the waiting room
nobody_joinedfailedIt was alone for 10 minutes
meeting_not_foundfailedA wrong or expired link, or a meeting that has ended or not started
not_accessiblefailedThe meeting needs a sign-in, a registration or a password, is locked or full, or is limited to an organization
busyfailedNo meeting bot was free after ten tries
insufficient_balance, too_many_calls, price_not_set, tier_unavailable, character_deleted, org_disabled, test_limit, not_startedfailedIts call couldn't start when it was due
interrupted, errorfailedWe lost it, or it failed on our side

A meeting that never got in is never charged. More reasons may come: treat one you don't know by its status.

Follow, list, move and call off

GET/v1/meetings/{meeting_id}

One meeting, brought up to date first. A meeting of your key's other env is 404 not_found, as if it didn't exist.

GET/v1/meetings

Your key's env's meetings, newest first: { "data": [...], "has_more": false }. A live key lists the live meetings (made through /v1 or the dashboard), a test key the test ones. limit is 1 to 100 (50 by default), before a created_unix to page back from, and status one of the five statuses to filter by (another is 400 invalid_field).

PATCH/v1/meetings/{meeting_id}

Moves a meeting to another time, or renames its bot, while it still waits here: before its bot is sent. It keeps its id. Send one or both:

Prop

Type

A field left out stays as it was; any other field is 400 invalid_field. It answers the meeting. Refused with 409:

CodeWhy
meeting_dispatchedThe bot is already on its way. Call it off and send a new one
meeting_overIt's over or called off
meeting_existsThe new time clashes with another meeting of the character at that link (error.meeting_id)
DELETE/v1/meetings/{meeting_id}

Calls it off before it joins, or takes the bot out of the meeting. It answers the meeting; one that is already over comes back as it is. As with reading one, a meeting of the other env is 404.

Webhooks

Each is sent once per meeting, to the endpoints of the meeting's env. data is the meeting, without error.

EventWhen
meeting.waiting_roomThe bot is waiting to be let in (waiting_room: true): the first time it waits
meeting.joinedThe bot got in (status: "in_meeting")
meeting.endedIt was in and left, or it was called off (status: "ended", with reason)
meeting.failedIt never got in (status: "failed", with reason). Never sent together with meeting.ended

The meeting's call sends its own call.started and call.ended. See Webhooks.

Test keys

A test key sends a real bot into the meeting, for 60 seconds at most, and counts against its 20 calls a day. It is never charged.

On this page