On-device AI notes for developers - transcribe locally, send only what you choose
Engineering and product teams often want AI summaries without turning every technical conversation into a vendor-hosted meeting record. On-device transcription handles the call locally, then lets you send only the passages worth processing to the BYOK provider of your choice.
When to keep transcription on-device
The full transcript starts on the device. Architecture details, customer names, roadmap discussion: all of it stays on the device until you decide a specific excerpt is worth sending out. BYOK stays optional, and when you do use it, only the selected text is governed by your provider's terms.
This is the right call when the meeting itself shouldn't be a cloud record, but a tightly-scoped excerpt (a summary, an issue list, a rewrite) needs the quality only a hosted model can deliver.
How it works
- Record the standup, design review, or product discussion on-device.
- Review the transcript locally before sending anything out.
- Select only the passage that needs a summary, issue extraction, or rewrite.
- Send that selection through your chosen BYOK provider.
- Keep the full transcript local unless your team has approved an export or cloud-provider transcript.
How it compares
| Option | Model choice | Data posture |
|---|---|---|
| Cloud meeting bots | Vendor-selected models | Full meeting record handled by the vendor |
| General AI chats | User-selected prompts | User pastes content into a provider session |
| jotty.pro on-device with BYOK | User-selected providers for selected text | Full transcript starts local; excerpts leave only when chosen |
Honest answers
Does BYOK undo the on-device benefit?
Only for the specific text you choose to send. The full transcript stays local; the excerpt you send to a provider is governed by that provider's terms. You can think of on-device + BYOK as "I want hosted-model quality for one paragraph, not for the whole hour."
Why not just use a normal meeting bot?
A meeting bot is fine when the vendor and data path are already approved. The on-device path fits teams that want a chance to review before any external processing happens, especially in the early stages of evaluating an AI workflow.
Can a cloud provider be used for team workflows?
Yes, but only after your team has reviewed the cloud transcript terms and decided that the content type belongs in a cloud workflow. Don't treat the configuration toggle as the approval decision.
If you want provider-grade speaker diarization or the freshest model versions on long technical calls, the cloud-provider version for developer notes routes both transcription and summary through your key.