Bring your own keys

By default the hosted API runs the model, the live transcription and the voice behind every bot on Meeting BaaS provider accounts, and that provider usage is billed to you alongside the bot's meeting time (see what provider usage costs). If you would rather use your own provider accounts — for billing, rate limits, data-processing agreements, or a model the server does not have a key for — pass them per bot in credentials. One limit to weigh first, especially for a data-processing agreement: the keys live only in the memory of the running API, so if it restarts while the bot is in a call, the bot is respawned on the server's own keys.

{
  "meeting_url": "https://meet.google.com/abc-defg-hij",
  "bot_name": "Analyst",
  "prompt": "...",
  "llm_provider": "anthropic",
  "llm_model": "claude-opus-4-7",
  "credentials": {
    "llm_api_key": "sk-ant-...",
    "stt_api_key": "<Deepgram key>",
    "tts_api_key": "<Cartesia key>"
  }
}

Key	Provider	Used for
`llm_api_key`	the one in `llm_provider`: OpenAI, Anthropic or Z.ai	the conversation
`stt_api_key`	Deepgram	hearing the room
`tts_api_key`	Cartesia	the bot's voice

Every key is optional. Whatever you leave out falls back to the server's own key, so you can bring only the LLM key and keep speech on MeetingBaas. A key you supply also satisfies the provider check: llm_provider: "zai" works with your Z.ai key even when the server has none.

What happens to the keys

They are used only by the pipeline process of that one bot, passed to it through its environment, and discarded when the bot leaves.
They are never written to disk, never logged, and never returned by any endpoint.
Validation errors (422) do not echo the request body.

What provider usage costs

A bot on Meeting BaaS keys is billed for what the providers charged: speech-to-text per second of meeting audio, text-to-speech per character spoken, the language model per token. A bot on your own keys is billed only for its meeting time; the providers bill you directly. You can mix: your LLM key with Meeting BaaS speech, for example.

Every bot keeps a usage report, live during the call and final afterwards:

curl -s https://speaking.meetingbaas.com/bots/<bot_id>/usage \
  -H "x-meeting-baas-api-key: $KEY"

{
  "stt_audio_seconds": 1812.4,
  "tts_characters": 6950,
  "llm_turns": 14,
  "llm_prompt_tokens": 61200,
  "llm_cached_prompt_tokens": 48800,
  "llm_completion_tokens": 2100,
  "llm_provider": "openai",
  "llm_model": "gpt-5.4-mini",
  "key_sources": { "llm": "caller", "stt": "platform", "tts": "platform" },
  "duration_seconds": 1830.2
}

key_sources says, per provider, whether that bot ran on a Meeting BaaS key (platform, billed to you) or on one you supplied (caller, billed by the provider). The same report is delivered in the speaking_bot.ended event when you set a callback_url. Current provider rates are on the pricing page.

What stays on the server's keys

The MeetingBaas side (joining, recording, the final transcript) always runs on the MeetingBaas account whose key is in x-meeting-baas-api-key. Deriving a persona from a prompt and generating persona images also use the server's own OpenAI and Replicate accounts.

What happens to the keys

What provider usage costs

What stays on the server's keys

On this page