Skip to main content
set a default in settings or pick per meeting before recording. transcription, ai insights, and filler word detection all adapt to the selected language. the table describes deepgram support. on-device transcription on macos supports a smaller set: english, spanish, french, german, portuguese and italian, plus dutch and russian with weaker results in our tests. greek, polish and swedish need deepgram.

providers

  • transcription - deepgram nova-3
  • managed ai insights - gpt-6-luna for speaker naming, catch-up, playbook topic extraction, template drafting and full and incremental standard summaries; pinned gpt-5.4-mini-2026-03-17 for sales, questions, template filling, grounded playbook answers and investigations
  • byok ai insights - pinned gpt-5-mini-2025-08-07, except investigations use pinned gpt-5.4-mini-2026-03-17
  • filler word detection - local, rule-based with language-specific defaults
from 2026-11-01, managed speaker naming, catch-up, playbook topic suggestions, template drafting and standard summaries use gpt-6-luna. this includes live updates and saved-meeting summaries on existing managed clients. sales, questions, template filling, grounded playbook answers, investigations and byok keep their existing models. no app update is needed for this model change.
from 3.7.0: byok summaries, speaker naming, catch-up, playbook topic suggestions and template drafting use gpt-6-luna with reasoning off. sales, questions, template filling, grounded playbook answers and investigations use pinned gpt-5.4-mini-2026-03-17 with low reasoning. your openai account must have access to the selected models. miniti shows provider errors without silently switching models. this requires an app update; linux releases separately.

picking language mid-meeting

language is set before recording starts. you cannot change it while recording - start a new session if you need to.