C0:DIC.
Speak it into existence.
Local and cloud transcription with commands, swappable pipelines, a self-learning coding dictionary, native text injection, and a floating bar that snaps to any screen edge.
Capabilities / shipped today
What it does.
Every claim below is backed by working code in the current build.
Push-to-talk on Ctrl+Shift+D
Hold to record, release to transcribe — or tap to toggle, your call.
globalShortcutService registers the OS-level binding. Toggle the press-to-hold ↔ tap-to-toggle behavior in settings.
Gemma on-device
A local Gemma pipeline can clean up stumbles into clearer prose when you choose on-device processing.
The gemma-local sidecar (LiteRT-LM runtime, VAD-segmented streaming) lives on your machine when local mode is selected.
Swappable transcription pipelines
Groq Whisper, Google Chirp, OpenAI Whisper API, OpenAI Realtime, BetterWispr, Faster-Whisper, Whisper Cloud, offline whisper.cpp, and local Gemma options.
pipelineAccessService.ts enumerates available backends. Switch between them per session, no relaunch required.
Domain dictionaries
Cardiology, anesthesiology, pharma, biotech, architecture, finance, devops, and programming vocabulary with phonetic and fuzzy matching.
dictionaryTrie.ts is the engine. Custom terms, hotwords, and corrections stay close to the dictation workflow.
Floating bar overlay
A minimal always-on dock that snaps to any screen edge, shows live amplitude, and exposes record / transcribe / chat.
floatingBarRenderer.tsx — separate Electron window, drag-to-snap behavior, persists position across launches.
Native text injection on Windows
SendInput injects transcribed text into Word, Notepad, web forms — anywhere keyboard input works.
injectionService.ts uses koffi to call user32.dll SendInput on Windows. Linux/macOS fall back to clipboard + Ctrl+V via an isolated ClipboardSandbox.
Pipeline leaderboard
Live latency comparison across available backends, so you can choose the route that fits the session.
PipelineLeaderboard.tsx renders first-word latency and total transcription time per pipeline, per session.
Custom abbreviations & hotwords
"EKG" → electrocardiogram. Inject medications, procedures, project codes that the model would otherwise miss.
hotwordProcessor.ts + abbreviationExpander.ts run before injection — your custom terms always win.
Audio + visual feedback
Pulse animation while recording, suppressible UI sounds for start/stop, status timer in the toolbar.
UISoundService + TranscriptionStudioToolbar. Mute the chimes if you're in a meeting.
vs the alternative
C0:DIC vs Dragon & Whisper CLI
- Eight backends, swap any time
- Built for medical, legal, dev jargon out of the box
- Native Windows injection — no Auto-Type hacks
- Floating bar that stays out of the way
Compare