AI Notes Guide

Turn meetings into useful notes, transcripts, and handoffs.

MediaSFU AI Notes attaches to a live room, captures the transcript pipeline, and produces readable post-session artifacts. Use it for meeting recaps, support handoffs, multilingual sessions, and notes-only rooms where translated playback is not needed.

Notes-onlyrun summaries without translated playback
Private firstdashboard access stays owner-scoped
MD + TXTnotes and transcript exports
Passcode shareonly when public access is enabled
Prove it in 2 minutesCreate a real transcript from a live room before you configure anything.
Includedwith team plansRides the translation pipeline — Pro adds ~100 translation minutes monthly
Use cases

Use AI Notes when the value is what happens after the call.

Meeting recaps

Turn meetings into decisions, action items, and readable summaries without forcing everyone to replay the recording.

Support and sales handoffs

Give the next teammate a clean package after a call: transcript, summary, open questions, and follow-up context.

Multilingual sessions

Run notes alongside translated rooms, or keep the room text-only when all you need is transcript and summary output.

Flow

From live audio to shareable files.

1

Configure providers

Set STT and LLM provider choices on the translation config. Add TTS if the room also needs translated voice playback.

2

Run the live session

MediaSFU captures transcript entries during the room and keeps the note pipeline attached to that session.

3

Generate artifacts

After the session finalizes, notes and transcript files become available when generation and persistence are enabled.

4

Review or export

Owners use Dashboard for review. Public viewers only see shared files through the passcode-protected AI Notes page.

Evaluate transcription well

Before you judge the notes, test the speech layer intentionally.

Set the spoken language

Use an explicit language when you know it. Auto-detection is convenient for broad demos, but it is not the strongest benchmark for every language or dialect.

Compare STT providers

Speech models have different strengths by language, dialect, region, and acoustic conditions. Compare candidates such as Sarvam, AssemblyAI, and Speechmatics with the same source audio; these examples are not a universal ranking.

Use representative speakers

Evaluate with the microphones, accents, room noise, speaking pace, and vocabulary your production sessions will actually contain.

Inspect transcript before summary

Treat speech recognition and note generation as separate layers. First verify the transcript, then tune the LLM prompt and summary structure.

Demo rooms favor broad defaults so more people can try the flow quickly.

For a production accuracy decision, configure the expected spoken language and compare providers using the same representative recording. No single STT provider is best for every language, dialect, accent, region, or recording environment.

Tune the speech pipeline
Setup requirements

Three decisions determine whether notes will show up.

Attach a translation config

AI Notes use the same real-time speech pipeline as translation. The room needs a usable translationConfigNickName, even when participants only need notes.

Enable notes for the room

Use enableAiNotes for normal rooms, or aiNotesOnly when you want transcription and summaries without translated audio playback.

Choose the review path

Keep notes private in the owner dashboard, or enable public access when outside viewers should retrieve safe exports with room name and passcode.

Developer shape

Notes-only room example

This is the common payload when you want transcripts and AI notes without translated playback for participants.

{
  "supportTranslation": true,
  "translationConfigNickName": "meeting-notes",
  "translationOutputMode": "text-only",
  "enableAiNotes": true,
  "aiNotesOnly": true
}
Implementation details

The important rules, without making you read the whole API reference.

Room flags

supportTranslation, translationConfigNickName, enableAiNotes, aiNotesOnly, and translationOutputMode decide what runs for a room.

Reusable defaults

aiNoteTakerConfig belongs on the translation config when you want reusable note behavior across rooms.

Notes-only mode

aiNotesOnly keeps the translation runtime available for transcript and summary generation, but normalizes output to text-only.

Public sharing

Public access is optional. When enabled, the safe retrieval page still requires the room name and passcode.

Outputs

What users actually get after the meeting.

Notes MD

Markdown summary output for meeting recap, decisions, action items, and follow-up sections.

Transcript MD

Readable markdown transcript for review, internal documentation, and handoff workflows.

Transcript TXT

Plain-text transcript for search, archiving, downstream processing, or custom workflows.

Troubleshooting

When notes are missing, check the pipeline in this order.

No notes appear after a meeting

Confirm the room used a valid translationConfigNickName and had enableAiNotes or aiNotesOnly enabled before the session started.

Transcript exists but summary is missing

Download the transcript first, then check whether generateNotes and persistArtifacts are enabled on the note-taker configuration.

Public viewer cannot access files

Public access must be enabled by the owner, and the viewer needs the exact room name plus passcode. Session ID is only needed for repeated rooms.

Notes are too short or noisy

Inspect the transcript first. If recognition is weak, set the language explicitly, compare STT providers, and test representative audio. If the transcript is sound, tune transcript caps, chunk duration, the LLM credential, and the aiNoteTakerConfig prompt.

Need the exact contract?

Use the technical reference when you are wiring rooms or saved configs.

The reference covers REST routes, authenticated app configuration, aiNoteTakerConfig limits, notes-only normalization, and public sharing boundaries.