OwenFlow

A local, unlimited Wispr Flow clone for Windows. Hold a hotkey, speak, release — your words are transcribed locally with Whisper (no usage limits, no cloud STT) and typed into whatever app has focus.

How it works

Hold Right Ctrl ──► mic records ──► release ──► faster-whisper (local sidecar)
        ──► optional MiniMax cleanup (fillers/punctuation) ──► pasted into focused app

• Electron app (app/) — tray icon, floating recording pill, settings + history UI, global push-to-talk hook, clipboard-paste injection.
• Python sidecar (sidecar/) — FastAPI + faster-whisper on 127.0.0.1:8484, model stays warm between dictations.

Install (packaged app)

cd app
npm run build:win        # produces dist\owenflow-<version>-setup.exe

Run the setup exe — it installs OwenFlow with a desktop + Start-menu shortcut and bundles the sidecar. Python 3.13 with the sidecar deps must still be installed (pip install -r sidecar\requirements.txt once via py -3.13 -m pip); the app launches the sidecar with py -3.13.

Setup (dev mode)

# 1. Sidecar (one-time; first run downloads the Whisper model ~460MB)
cd sidecar
pip install -r requirements.txt
python test_sidecar.py   # optional sanity check

# 2. App
cd ..\app
npm install
npm run dev              # or: npm run build && npm start

The app spawns the sidecar automatically on launch. Tray icon → Settings to configure.

Usage

• Hold Right Ctrl (default) and speak; release to transcribe and paste. Configurable hotkey + a press-to-toggle mode in Settings.
• Cleanup (optional): add your MiniMax API key in Settings to strip filler words and fix punctuation. If it's off, slow, or fails, the raw transcript is pasted — dictation never blocks on AI. Refinement defaults to Groq (llama-3.3-70b-versatile, sub-second); add a Groq API key in Settings → Modes. MiniMax stays selectable as a slower max-polish option, and a Test & compare button times both providers head-to-head.
• Dictionary: one entry per line in Settings. Plain words bias recognition (e.g. Cresio, Fluxboard, ZEAL); wrong=>right entries are find/replace fixes.
• History: tray → History for recent transcripts with copy buttons.
• Snippets: say a saved trigger ("sign off email") and its expansion is pasted verbatim (skips AI cleanup) — configure in Settings → Dictionary.
• Translate mode: a flow mode that transcribes any spoken language and pastes a translation into your target language (set the target in Settings).
• Sessions: pick an active session from the tray (e.g. client / notes); its tone is applied automatically and each dictation is tagged with the session label.
• App profiles: when enabled (Settings → Apps), OwenFlow detects the focused app and reshapes output per-app — code-friendly in editors/terminals, prose in browsers, formal in mail — with optional per-app flow mode, transforms (strip trailing period, no auto-capitalize, single line), find/replace, and a prompt rule. Off by default.
• Auto-learning dictionary: edit a transcript in History and click Learn — OwenFlow diffs your fix against the original and proposes a wrong=>right entry to add, so the same mistake self-corrects next time.
• Never lose a dictation: if the Whisper sidecar is cold or busy when you finish speaking, the audio is queued and retried in the background; the recovered transcript lands in History (tagged recovered) with a notification you can click to copy — it's never pasted late into the wrong app.
• Daily digest: a once-a-day tray notification summarizing what you dictated (count, words, time saved vs typing), with an optional AI theme summary. Configure the time/toggles in Settings → General, or trigger it anytime from the tray's Today's digest.
• Command mode (speak-to-act): enable a second hotkey in Settings → General. Select some text, hold the command hotkey and speak an instruction ("make this a bullet list", "rewrite formally") — it edits your selection via the LLM and pastes the result; with nothing selected it generates from the instruction. Off by default.
• ZEAL voice command: in command mode, prefix with "ZEAL …" (or "note …") and the instruction is sent to your ZEAL agent's /api/voice endpoint — ZEAL executes it (tasks, missions, kanban, etc.) and replies; the reply pops as a notification and, with Speak ZEAL replies on, is read aloud via the sidecar's edge-tts. Configure the endpoint + key in Settings → ZEAL.
• Continuous / draft mode: toggle it on in Settings → General, then just hold the dictation hotkey and keep talking — OwenFlow splits on your natural pauses and streams each segment in (transcribed + cleaned) while you continue, so long-form drafts flow in paragraph by paragraph. Release to finish; the whole session lands as one History entry. Off by default.

Customizing

Want to change…	Look in
Whisper model size / language	Settings UI, or OWENFLOW_MODEL env for the sidecar
Pill look & feel	app/src/renderer/pill/
Refinement provider (Groq default / MiniMax)	Settings → Modes, or app/src/main/cleanup.ts
Injection method	app/src/main/injector.ts
Hotkey behavior	app/src/main/hotkey.ts

Design spec: docs/superpowers/specs/2026-06-12-owenflow-design.md.