Models & Compatibility

With local models, dictation and correction complete fully without an internet connection. Whispree tells you which models your Mac can run comfortably via the ‘Can I Run’ grade.

Local model registry

Download and manage supported models under Settings → Models.

STT: WhisperKit Large V3 Turbo (~1.5 GB), Qwen3-ASR-1.7B-8bit (~1.0 GB).
Correction (text): Qwen3 (1.7B/4B default/8B), Qwen3 Coder 30B, Gemma 4 (2B/4B/26B MoE/31B), SuperGemma4 26B, GLM-4.7 Flash, and more.
Correction (vision): Qwen3-VL-4B.

Each model can be downloaded/canceled/deleted inline, with progress (bytes · %) and an “in use” badge shown. If another download is in progress, it queues up, and you can retry on failure.

‘Can I Run’ compatibility

By detecting your Mac’s chip, RAM, memory bandwidth, and GPU cores, Whispree computes and shows the following per model (canirun.ai style):

RAM usage (%), estimated speed (tok/s), quality score (0–100),
and an at-a-glance compatibility grade (6 tiers, e.g. RUNS GREAT … TOO HEAVY).

Cloud models (OpenAI/Groq/Claude) always show as RUNS GREAT. These metrics are surfaced throughout the STT, LLM, and Models tabs.

Storage location

Models are stored in ~/.cache/huggingface/hub/. You can check this directly via “Open in Finder” in Settings.

Other convenience settings

Features you can toggle under Settings → General:

Setting	What it does	Default
Pause music while recording	Pauses currently playing media (Now Playing sources such as Apple Music, Spotify, YouTube) when recording starts, then resumes when it ends.	On
Audio input channel	Selects which mic channel to use on a multi-channel input device (0 = auto downmix).	0 (auto)
Launch at login	Starts Whispree automatically at macOS login.	Off

Auto-update

Whispree checks for updates automatically via Sparkle. You can also check manually with “Check for Updates…” in the app menu.