Meetings, recorded and recapped on your machine

Korely records both microphone and system audio on macOS, Windows, and Linux, so you capture everything on the call, not just your side. Hit the record button in any note or in the floating mini recorder. The audio file attaches to the note as you go and the live transcript fills in next to it.

On Free, the live transcript is produced by Whisper running on your own machine. Pro switches to Deepgram for cloud streaming with speaker diarization, punctuation, and named entity capitalisation. That last one matters when you can't tell from context who said what.

Already recorded the call elsewhere? Drag any .mp3, .m4a, .wav, .mp4, or .mov file into Korely. The same pipeline runs: transcript, then diarization, then recap. The original media stays in your vault next to the note, ready to replay whenever you need to double check what someone actually said.

For Zoom and Teams users this replaces the "wait for the cloud to send you the recording" detour. Export from the platform, drop into Korely, get a searchable, AI readable transcript in minutes.

Long meetings produce long transcripts that nobody actually re-reads. Pro Korely takes the transcript and turns it into a recap you will actually open. It picks the right template depending on the meeting type, whether it's a sales call, an interview, a one to one, a standup, or just a generic meeting, then returns a sectioned summary:

• Decisions taken: explicit and implicit, with timestamps
• Action items: owner, verb, and due date when stated
• Open questions: anything left unresolved at the end of the call
• Notable quotes: verbatim moments worth remembering

Behind the scenes Korely uses Gemini Flash for the recap. It's fast, affordable, and good at this specific job. The transcript itself stays in your vault, like every other note. Only the text is sent to the model.

Deepgram returns generic labels: Speaker 0, Speaker 1, and so on. That's fine until you re-read the transcript a week later and have no idea who said what. Pro Korely runs a second small model on top, Groq's Llama 3.3 70B with sub second latency, and infers real names from self introductions such as "Hi, I'm John" or from a list of participants you pass in.

If you list the attendees before the call ("John Smith (CEO), Sarah Chen (CMO), me"), the recap maps the diarisation labels to those names. The transcript reads like a play, not a tech support log.

Once transcribed, a meeting is just a Markdown note in your vault. The transcript is the body, the recap sits at the top, the audio file lives in attachments/, and the entity graph wires it to every other note mentioning the same people, project, or topic. More on the knowledge graph →

That also means: search ("what did the customer say about pricing?"), MCP access from Claude or Cursor, drag and drop into other notes, and folder organisation all work on meetings the same way they work on regular notes. No separate "meetings module" with its own quirks.