Sample
See a finished report
A scripted demo meeting, three named speakers, and the report KaigiFlow wrote from it: decisions, actions with owners and dates, risks and open questions.
Open the sample report
The meeting ends. The record gets better.
In short
Live text is the first draft. Afterwards you can replay any sentence, confirm who spoke, improve the transcription and translation, and export the revision you trust. Voice matching is assistive, not biometric identification.
One meeting — before, during, and after
KaigiFlow carries a single continuous record from setup to export.
Before
Set the input language, pick up to three outputs, add participants, and choose the capture source.
During
Read live transcription and translation with stable speaker labels. Pause when you need to. The recording stays on your device.
After
Replay any sentence, confirm speakers, retranscribe, retranslate, rename, and export the revision you choose.
Know who spoke — without guessing
KaigiFlow separates speakers as the meeting happens and helps you confirm them afterward. Uncertain speakers stay unresolved until you decide.
- Designed for meetings of two to eight speakers.
- Import participant lists from text, images, or PDFs.
- Confirm speakers manually, or save voice samples for people you meet often.*
- Automatic remapping is confidence-aware and always reviewable.
Save someone's voice sample only with their agreement.
Replay the exact sentence. Keep the revision you trust.
Select any line in the transcript and the audio jumps to that moment.
- Playback follows the active sentence as you listen.
- Switch between the live and refined transcript and translation revisions.
- Exports reflect the revisions you selected.
What we measured
Our own tests, on public recordings with human transcripts. Lower error rates are better. Each figure carries how it was tested and when.
Re-read afterwards: up to 4× fewer errors in Cantonese, about 2× in Thai[4]
Results of the after-meeting re-read, compared with standard cloud speech recognition. In English and Mandarin meetings the re-read has 14% to 19% fewer errors. Lower is better. Claimed only for the languages shown.
Read Cantonese
- KaigiFlow
- 10.4%
- Standard cloud speech recognition
- 40.8%
Thai
- KaigiFlow
- 10.7%
- Standard cloud speech recognition
- 24.2%
English meetings
- KaigiFlow
- 24.8%
- Standard cloud speech recognition
- 30.7%
Mandarin meetings
- KaigiFlow
- 39.3%
- Standard cloud speech recognition
- 45.7%
Tests and dates
- [4] After-meeting re-read of English and Mandarin meeting recordings and of Cantonese and Thai read speech, all with human transcripts, compared with standard cloud speech recognition. Measured with Sangei Cloud. Tested October 2026. How we measure
- [9] All tests used public recordings with human transcripts. No customer recordings were used. Results are from KaigiFlow’s own tests in October 2026 and describe those tests, not every meeting. How we measure
Questions people ask
Meeting capture, supported languages and what happens to your recordings.
Can KaigiFlow tell who is speaking?
KaigiFlow separates speakers during a meeting and can use voice samples you save for people you meet regularly. Speaker labels are assistive, not biometric identification. Uncertain speakers remain unresolved until you confirm them. Review speaker assignments after the meeting instead of treating every automatically assigned name as a confirmed identity.*
Save someone's voice sample only with their agreement.
Can I delete my meeting recordings?
Yes. You can delete meetings and voice profiles in the app. Recordings and other local meeting information remain on your device unless you export them or use the cloud features described in the Privacy Policy. Exported files are yours to manage separately. Cloud processing follows the retention periods in that policy.
Bring clarity to your next multilingual meeting.
See the beta details for Mac, iPhone and iPad — status, compatibility, and install notes for the private beta.
Join the private beta