Hold a hotkey. Talk. Release. The transcript is polished into structured text and pasted at your cursor in any app. Run it offline on your Mac for free, or use our hosted service for zero local memory and faster latency.
Hold ⌥ Option · the floating pill shows your voice in real time
Runs entirely on your Mac. No accounts, no subscriptions, no data leaves your machine.
Cloud transcription with our keys. Zero local memory. WisprFlow-tier latency. Free tier: 2,000 words/month.
Same desktop app for both. Switch from local to Pro (or back) anytime from the menu bar.
Same flow for Local and Pro. Different inference, identical UX.
Right Option key by default (configurable). The floating pill appears at the bottom of your screen with 16 bars rippling to your voice.
Whisper transcribes (locally on MLX or remotely on Groq), an LLM polishes the structure — ordinals become numbered lists, fillers get removed, self-corrections get applied.
The polished text shows up exactly where you were typing — Slack, VS Code, Notes, Gmail, anywhere with a text field.
When to pick which.
| Susurro Local | Susurro Pro | WisprFlow | |
|---|---|---|---|
| Price | Free, MIT | $10/mo (2k free) | $15/mo |
| Local RAM | ~3 GB | 0 GB | 0 GB |
| Latency (5s clip, M3 Pro) | ~1.5 s | ~0.7 s | ~1–3 s |
| Word cap (paid tier) | — | 100k/mo (published) | "Fair use" (vague) |
| Offline | Yes | No | No |
| Audio leaves your Mac | Never | Yes (Groq · no training) | Yes (proprietary) |
| Open source | Yes, MIT | Yes, MIT (client) | Closed |
| Hotkey + waveform pill | Yes | Yes | Yes |
| Smart formatting (lists, fillers, backtrack) | Yes (local Llama 3B) | Yes (Llama 70B) | Yes |
macOS 13+ on Apple Silicon. One shell command.
curl -fsSL https://raw.githubusercontent.com/danilobrando/susurro/main/install.sh | bash
The installer checks for Apple Silicon, installs Python deps via pipx, downloads Whisper + Llama 3B (one-time, ~5 GB), and registers the susurro command. After install, you'll grant 3 macOS permissions and hold ⌥ to talk.
Yes — same app, single click in the menu bar. Susurro Pro saves a token in ~/.susurro/auth.json; signing out goes back to Local. The hotkey, waveform pill, polish prompt — all identical.
Most paid dictation tools say "unlimited with fair use" and throttle you privately. We publish the cap: 100,000 words/month before a soft warning, 110,000 hard cap with a 24h cooldown. At average ~50 words per dictation that's ~2,000 dictations/month — more than even heavy daily users. If you need more, contact us.
Audio is streamed to Groq for Whisper transcription + Llama polish, then dropped. Nothing persisted on our servers — we only store the final word count for usage metering. Groq's TOS states they don't train on API data. If you need zero-network privacy, use Susurro Local.
The local backend uses Apple's MLX framework for Whisper and Llama, which only runs on M-series chips. Susurro Pro doesn't have this restriction — once we ship the Intel/Windows desktop client, you'll be able to use Pro from anywhere.
An LLM looks at the raw Whisper output and applies four kinds of edits: ordinals like "primero, segundo, tercero" become a markdown numbered list; obvious fillers (eh, mmm, you know, um, uh) get removed; self-corrections ("Pedro, eh, Pablo digo") collapse to the final intent; long monologues get paragraph breaks. It explicitly refuses to paraphrase or change the speaker's words.
Email hi@susurro.live. We refund first-month cancellations within 7 days, no questions asked.
Yes — that's Susurro Local's "advanced" mode. Set SUSURRO_GROQ_API_KEY in your environment, change STT_BACKEND="groq" in susurro/config.py, and the app proxies through your own key. Free, no Susurro Pro account needed.