Configure the transcription provider for a Recall.ai recording (real-time streaming vs async, Recall vs third-party)

Documented steps

1. Decide streaming vs async first: streaming providers emit text during the meeting (live agents/copilots); async providers process after the recording ends and are generally cheaper and more accurate. Both are set at recording_config.transcript.provider.<provider_key> in POST /bot.
2. Recall's own streaming provider: {"provider": {"recallai_streaming": {"mode": "prioritize_low_latency", "language_code": "en"}}}. mode 'prioritize_low_latency' is English-only with ~1-3s latency; mode 'prioritize_accuracy' supports 40+ languages and language_code 'auto' but is delayed several minutes. Ref: https://docs.recall.ai/docs/recallai-transcription
3. Recall's async provider: {"provider": {"recallai_async": {"language_code": "auto"}}}, supporting 40+ languages, key_terms custom vocabulary, spelling correction and profanity filtering. Ref: https://docs.recall.ai/docs/ai-transcription
4. Third-party streaming provider keys documented by Recall include assembly_ai_v3_streaming, deepgram_streaming, elevenlabs_streaming, aws_transcribe_streaming, rev_streaming, speechmatics_streaming and gladia_v2_streaming. Ref: https://docs.recall.ai/docs/bot-real-time-transcription
5. Third-party async provider keys include assembly_ai_async (plus a chunked variant) and deepgram_async. Ref: https://docs.recall.ai/docs/assemblyai and https://docs.recall.ai/docs/deepgram
6. AssemblyAI example (async): {"assembly_ai_async": {"speech_models": ["universal-2"], "language_detection": true}}. Streaming: {"assembly_ai_v3_streaming": {"speech_model": "universal-streaming-multilingual", "language_detection": true, "format_turns": true}}.
7. Deepgram example (async or streaming): {"model": "nova-3", "language": "multi", "mip_opt_out": true}.
8. Third-party vendor API keys are NOT passed in the request body - store them in the Recall dashboard for the same region the bots run in, and re-enter them separately in every region you operate.
9. Verify the provider key is saved in the correct regional dashboard before going live; a missing key causes bot creation/transcription to fail for that provider.