Local AI Guide — On-Device Transcription & Privacy

Fuse Intelligence products run inference on your machine. VocalFuse is the first — a Windows 10+ desktop app with a local transcription model (~~57 MB), hold-to-speak dictation into any app, and Pro modes for AI note taking without uploading audio.

On-device only Offline dictation ~57 MB model Windows 10+

Local transcription model

VocalFuse downloads a local transcription model (~57 MB) on first launch into C:\VocalFuse\models\. The model stays on your machine and powers all three modes:

  • Transcribe — hold the pill, speak, release; text pastes at your cursor with professional auto-punctuation
  • Note Taking (Pro) — click start/stop; AI summary + transcript segments can sync to Account → Notes
  • Continuous (Pro) — background 60-second clip checks; bullets stay local under AppData

Audio is never sent to a cloud speech API. See the full VocalFuse documentation for install, product keys, and troubleshooting.

Why local AI for legal, medical & confidential work

Cloud transcription services store audio on remote servers and often require a meeting bot. VocalFuse keeps the trust boundary on your PC: microphone → local model → text at cursor or Account → Notes (text only). No Otter-style bot joins Zoom or Teams. Compare Otter alternatives and private AI meeting notes.

Performance & offline use

After the model downloads, hold-to-speak dictation works offline. You need HTTPS for license checks, first model download, Pro AI summaries, and optional note sync. Faster CPUs reduce “Working…” time between hold and paste; 16 GB RAM is recommended while the model is loaded.

If latency feels sluggish, close other heavy applications and confirm the model finished downloading before your first session. There is no Performance / CUDA card in Settings today.

Plans & product keys

Basic ($5/mo) — unlimited local dictation, pill widget, product key, updates. Pro ($10/mo) — adds Note Taking, Continuous, cloud notes toggle, and optional Gmail (notes_email_to). Subscribe on Pricing, copy your key from Product Keys, paste in the desktop installer or Settings → Product License.

Building your own integration

Use the license verification and notes APIs documented on this site to integrate with Fuse Intelligence from your own tooling. Questions? Ask in the community forum.

Local AI FAQ

Why use local AI transcription instead of cloud?

Local inference keeps audio on your PC, removes network latency for dictation, works offline after the model downloads, and avoids per-minute cloud billing. VocalFuse is built on this model.

How big is the VocalFuse transcription model?

About ~57 MB downloads on first launch into C:\VocalFuse\models\. It stays on your machine for Transcribe, Note Taking, and Continuous modes.

Does local AI mean nothing ever leaves my PC?

Audio never uploads for transcription. Pro Note Taking can upload text transcripts and AI summaries to Account → Notes when cloud notes are enabled — configurable in Account Settings.

Do I need a GPU for local transcription?

No GPU is required. VocalFuse runs on CPU. Faster machines reduce processing time; there is no GPU toggle in the current Settings panel.