Meeting Recording

EM Kit can record the audio of any meeting attached to a note, transcribe it, and turn the transcript into a summary. The recorder sits at the top of every 1:1, Daily Meeting, and Custom Note, so you can capture the conversation without leaving the page.

Recordings are stored locally next to your database, and you choose whether transcription stays on your machine (whisper.cpp) or runs in the cloud (AssemblyAI).

Where the recorder lives

The recorder is mounted on every note that benefits from a conversation:

• One-on-One notes.
• Daily Meeting notes.
• Custom Notes.

You’ll see a thin bar above the note content with a Record meeting button, the active engine label, and a quick toggle between audio sources.

Record a meeting

1. Open a note where you want to record the conversation.
2. Pick the audio source on the right of the bar — mic or mic+sys (see below).
3. Click Record meeting. A live waveform and elapsed timer appear.
4. Click Stop recording when the meeting ends, or Discard to throw the recording away.

After you stop, the audio is finalized to disk and queued for transcription. Progress is reported live — first an “uploading” phase, then “transcribing” with a percentage.

When transcription finishes, a card appears under the bar with the status, duration, file size, detected language, and an expandable transcript.

Audio sources

The toggle on the right of the recorder bar switches between two capture modes:

Source	What it captures	Notes
mic	Microphone only	Most reliable across operating systems. Best for in-person meetings or when remote participants are heard through your speakers via a headset mic.
mic+sys	Microphone + system audio	Captures both your voice and the audio coming out of your computer (Zoom, Meet, Teams, etc.). Requires extra OS permission — see below.

You can change the default audio source in Settings → Recording so new recordings start with your preferred mode.

Transcription engines

EM Kit ships with two transcribers. Pick one in Settings → Recording → Transcriber.

whisper.cpp (local)

Runs entirely on your machine. No audio leaves the device.

• Install the binary once with brew install whisper-cpp, or point Custom binary path at an existing install.
• Pick a model — larger models are more accurate but slower and use more disk. Models are downloaded on demand into userData/whisper/models/.
• The first transcription will fetch the selected model automatically; you can also click Download now from Settings to grab it ahead of time.

AssemblyAI (cloud)

Sends audio to AssemblyAI’s API and returns a transcript. Faster and often more accurate, but the recording leaves your device.

1. Create a free AssemblyAI account — new accounts include free credit (≈ 100 hours of transcription).
2. Open the API Keys page in their dashboard and copy your key.
3. Paste it into Settings → Recording → AssemblyAI API key and click Save.

Choose between Universal-3 Pro (most accurate, 6 languages) and Universal-2 (broadest language coverage, lowest cost) in the Model dropdown.

Caution

With AssemblyAI selected, every recording you transcribe is uploaded to their servers. Switch to whisper.cpp if you need fully local transcription.

Language

By default the engine auto-detects the spoken language. Set Language hint in Settings to a code like en, fr, or es to force a specific language — useful for short recordings where auto-detection can be noisy.

Permissions

The recorder needs OS-level permissions to access your microphone and (optionally) your screen audio. Settings → Recording → Permissions shows the current status and offers buttons to request access.

Microphone

Required for any recording. The first time you click Record meeting, the OS will prompt you to grant access. If you previously denied it, open System Settings → Privacy & Security → Microphone and enable EM Kit.

Screen & system audio recording (macOS only)

Required when the audio source is mic+sys. macOS only adds an app to the Screen Recording list once it has attempted a capture — so the first time, the recorder will trigger a capture, fail with a “permission denied” error, and open System Settings for you.

1. In Settings → Recording, click Trigger prompt under “Screen & system audio recording”.

2. Enable EM Kit in the Screen & System Audio Recording list.

   Note

   In dev mode the entry is labeled Electron, not “EM Kit”.

3. Quit EM Kit fully (⌘Q) and relaunch — macOS only re-reads the permission on launch.

If you change source to mic+sys and the permission is missing, the recorder will open System Settings automatically and show a banner reminding you to relaunch the app.

Summarize the transcript

Each completed recording card has a Summarize button. Click it to send the transcript to your configured AI provider and get a short summary of the meeting — decisions made, action items, blockers — that you can paste into the note.

The summary uses the same provider configured for the rest of EM Kit’s AI features. See the EM Copilot guide for setup.

Manage recordings

Every recording attached to a note is listed below the recorder bar with its status:

Status	Meaning
recording	Capture in progress.
captured	Audio is saved, waiting to be transcribed.
transcribing	The engine is running.
completed	Transcript is ready.
error	Transcription failed — click the retry icon to try again.
canceled	You discarded the recording before it finished.

From each card you can:

• Show transcript — expand the full text inline.
• Summarize — generate an AI summary (see above).
• Retry — re-run transcription if it errored or stalled. Useful after changing engine or model.
• Delete — remove both the database row and the audio file from disk.

Where files are stored

Recordings are local-first, just like the rest of EM Kit:

• Audio files live next to the database, under meeting-recordings/ in the active data directory.
• Transcripts, summaries, and metadata are stored in the SQLite database.
• Whisper models live under whisper/models/ in the same data directory.

See Data Storage for the exact path on your OS and how to back up or move your data.

Troubleshooting

”whisper.cpp executable not found”

Install it with brew install whisper-cpp, or set Custom binary path in Settings to the absolute path of an existing binary.

”Model not downloaded”

Click Download now in Settings → Recording → Transcription engine, or just start a recording — the model is fetched automatically on first use.

”macOS blocked screen capture”

The recorder needs Screen & System Audio Recording permission for mic+sys. Follow the steps under Permissions, making sure to quit and relaunch EM Kit after enabling the permission.

Recording finished but the transcript is empty

The engine didn’t find any speech in the audio. Check the audio source (you may have selected mic+sys but no system audio was playing), and that your microphone wasn’t muted. You can click the retry icon to re-run transcription on the same audio file.