The three paths in Meetingly
| Feature | What it processes | Where it runs |
|---|---|---|
| Local transcription | Microphone and/or system audio | On your computer after the speech model downloads |
| Cloud transcription | Selected live audio | Through the Meetingly relay to Deepgram, when selected on Unlimited |
| Hosted AI answers and reports | Questions, prompt context, transcript text, and screenshots you request | Through the Meetingly relay to the configured AI provider |
| Your own API provider | The context needed for your AI request | Directly from the app to your configured endpoint |
| Connected account sync | Meeting history, settings and prompts | Stored by the account service and available on your linked devices |
What local transcription gives you
The default engine uses an on-device speech model. You need an internet connection to obtain the model initially. Once it is present, the transcription process runs on your computer. Performance depends on the machine, the selected language, audio quality and other applications competing for CPU time.
Local recognition keeps the audio out of the cloud transcription path. It does not stop transcript text from being included in a hosted AI question or report. It also does not disable account sync. Review those settings separately if your goal is to keep meeting content on the device.
When cloud transcription is used
Unlimited includes an optional cloud speech engine. When you select it, audio from the chosen sources is streamed through the relay to Deepgram for recognition. The relay does not persist the audio stream. The provider handles it under its own policies, which you should review for sensitive work.
Cloud recognition needs a working network connection and consumes the plan’s cloud transcription allowance. Local recognition depends on the computer instead. Compare both on your own language and hardware before deciding; a faster or more accurate result is not guaranteed for every meeting.
Where AI answers and reports go
The hosted plan sends the relevant text and any screenshot you explicitly request through the Meetingly relay to the configured model service. Your own-key mode sends that material to the endpoint you select. An OpenAI-compatible endpoint may be a cloud service or a server on your own machine.
For a local answer path, run a local model and set Meetingly to its local endpoint. The Ollama setup guide shows the workflow. Text-only models can answer text questions, but screenshot analysis requires a model that accepts images. A local endpoint address alone does not prove that the model is local; check your model server’s configuration too.
Account sync is a separate choice
Connecting an account enables synchronization of meetings, prompts and supported settings. Disconnecting a device prevents future synchronization from that connection. It does not automatically erase meeting records already stored in the account. Use the account dashboard to delete those records, or delete the account if that is what you intend.
For a session whose content must stay on your computer, use local transcription and a local answer model, keep the account disconnected, and check the endpoint before the session. App updates and model downloads may still use the network; this is a content-flow configuration, not a claim that the application makes no network requests.
A practical pre-meeting check
- Confirm the transcription engine before pressing Listen.
- Confirm the answer endpoint and model in the own-key panel, or the connected hosted plan.
- Check whether your account is connected and whether older history is already synced.
- Make sure recording and data-sharing choices meet the rules of the meeting.
- Use a short non-sensitive sample to validate the setup.
See the privacy policy for the complete current description. The application’s source is available in the Meetingly repository.