01 / CAPTURE
Voice input
Shuo captures microphone audio only while you hold the push-to-talk shortcut, then closes that input when you release.
Boundary Begins at this Mac’s microphone
A voice keyboard for Mac
Shuo
Voice typing for the words
only your team knows.
Hold to speak, release to insert, and fix any miss in the floating bar.
Local mode keeps voice and text on your Mac—no account or telemetry.
The way your work actually sounds
“Deploy this API to staging, then open a PR.”
Correction loop
LIVE LOOP / 01—03
› Use local Wishper for this endpoint, then fall back to cloud AI. Use local Whisper for this endpoint, then fall back to cloud AI.
Use local Wishper for this endpoint, then fall back to cloud AI.
Use local Whisper for this endpoint, then fall back to cloud AI.
Put back in place
Signal path / 01—07
What happens before a sentence lands.
02 / PREPARE
Audio processing
Speech detection and quiet-input adaptation prepare a clean audio segment for transcription.
Signal Audio preparation runs locally
03 / CONTEXT
Context
Preferred terms, project vocabulary, and reusable prompts become bounded spelling hints; project source files and paths are not uploaded.
Context Only relevant hints are prepared
04 / TRANSCRIBE
AI inference
Your selected local or cloud speech model turns the prepared audio into an initial output. Only cloud mode sends this audio and its bounded hints.
Model Local-first, cloud optional
05 / REFINE
Post-processing
Only enabled steps run: text cleanup, script conversion, emoji, replacements, and optional AI retouching.
Control AI retouch sends only the current text and required prompt
06 / REVIEW
Human correction
Revise in the floating bar and safely return the text to the original app.
Human loop You keep the final say
07 / COMMIT
Final result
Confirmed text returns to the target app; linked audio and history remain reviewable or deletable.
Result Written back where you work
Real working language
Names
Team names · project names · product terms
Names from every team and writing system.
Projects
Qilin · Aranya · Project Feishu
Supply internal names as context.
Engineering
RAG · pgvector · eval · SwiftUI
Give acronyms and code terms explicit spelling hints.
Rules and voice edits
YOU SAY
The Qillen launch is ready. Party popper emoji
SHUO TYPES
The Qilin launch is ready. 🎉
FIRST PASS
Ask the platform team to review the Qilin migration after stand-up.
YOU SAY
“Edit the last sentence. Change the reviewer to the release team and schedule the review before stand-up.”
RESULT
Ask the release team to review the Qilin migration before stand-up.
A corpus that stays yours
Every correction leaves a useful trace.
Audio, first-pass output, and accepted corrections stay linked on your Mac until deletion—an inspectable corpus for future personal models. Shuo does not upload or train on it; any future use requires your opt-in.
Original audio
First-pass output
Accepted correction
Replay, compare, delete
Future training is opt-in
A legible data boundary
Use local transcription with cloud AI off, and your voice and text stay on your Mac.
01 / Local
Settings, audio, history, metrics, project indexes, and corrections stay local—no account, telemetry or behavioral analytics sent to Shuo, and no ads.
02 / Cloud
Cloud transcription receives the current task's audio, settings, enabled context, and bounded spelling hints—not project source or paths.
03 / Control
API keys live in Keychain. Correction learning is off by default; each pattern must be enabled and pass safety gates. Future training or upload also requires explicit opt-in.
Download
Make the words that matter easier to say.
For macOS 14+. Local setup needs only a model download; cloud transcription is optional.