Agent Wispr
AI-powered voice dictation that types anywhere on your desktop
Press a hotkey. Speak. Your words appear wherever your cursor is — 100% local, no cloud, no subscription.
A lightweight floating widget that sits on top of whatever you're working in. Hold a hotkey, speak your prompt or message, and the transcribed text is pasted directly into the focused window — no switching tabs, no copy-pasting. Built for developers who spend their day talking to AI agents.
Privacy-first by design
Your voice never leaves your machine.
Most dictation tools send your audio to a cloud server. Agent Wispr runs entirely on your device — no microphone access by a third party, no recordings stored remotely, no internet required after first setup.
Features
100% Local AI Transcription
Powered by OpenAI Whisper running entirely on your machine. Your audio never leaves your device — no cloud, no subscription, no privacy trade-off.
Types Anywhere
Injects transcribed text directly into whatever window is focused — code editors, browsers, terminals, chat apps, email. No copy-paste required.
Toggle & Push-to-Talk Modes
Toggle mode: press once to start, press again to stop. Push-to-talk: hold the hotkey while speaking. Choose what fits your workflow.
Word Correction Dictionary
Teach the app your vocabulary once. Right-click any transcript entry to add a correction — it applies to all future and past transcriptions automatically.
Model Selection
Choose from Whisper tiny (fast, 75 MB) through large-v3 (most accurate, 3 GB). Swap models in Settings without reinstalling.
Full Transcription History
Every dictation is saved with timestamp and word count. Search, re-inject, export to CSV, or delete individual entries from the history panel.
Stats & Usage Tracking
See words dictated, audio recorded, transcription count, and average latency — per session and all-time. Sparkline trend for the last 7 days.
Cross-Platform
Works on Windows, macOS, and Linux. One purchase covers all three platforms. Native text injection on each OS.
Stays Out of the Way
Ultra-compact floating widget (95px collapsed). Hides from the taskbar and Dock. Auto-expands waveform during recording, snaps back when done.
Who It's Built For
Software Developers
- Dictate code comments, commit messages, and terminal commands without breaking flow
- Teach it your stack's vocabulary — library names, variable names, project-specific terms
- Corrections apply retroactively so your history stays clean
Privacy-Conscious Users
- Your audio is processed on your own hardware — no audio sent to any server
- Works fully offline after the initial model download
- No cloud account, no API keys, no usage logs
Accessibility & RSI Users
- Reduce keyboard usage throughout your workday
- Push-to-talk mode prevents accidental activations
- Ultra-compact widget stays out of your way when not recording
System Requirements
- Platforms: Windows 10/11, macOS, Linux
- Installation: No administrator rights required
- First run: Downloads the Whisper model (~75 MB–3 GB depending on model size) automatically
- GPU: Optional — CUDA auto-detected for faster transcription; CPU fallback included
FAQ
Does it work without an internet connection?
Yes. Transcription runs entirely on your machine using the Whisper model you've downloaded. Once the model is cached, no network connection is required.
How accurate is the transcription?
Accuracy depends on the model size and your accent. The large-v3 model is substantially more accurate on technical vocabulary and general English than the smaller models, at the cost of more VRAM and latency. The base model is fast and works well for clear speech in quiet environments. The word correction dictionary helps address jargon and project-specific terms regardless of model size.
What's the difference between Free and Pro?
Free includes the base Whisper model, 7-day rolling history, and in-session word corrections. Pro ($29.99 one-time) unlocks all models (tiny through large-v3), unlimited history, dictionary persistence across sessions, full stats export, and priority support.
What do I get when I buy?
You get compiled installers for all 3 platforms: Windows (.exe), macOS (.dmg), and Linux (.AppImage). No Python installation required — just download and run.
Does it work with CUDA / my GPU?
Yes. On Windows and Linux with an NVIDIA GPU, the app automatically detects CUDA and uses it for significantly faster transcription. CPU fallback is automatic if CUDA is unavailable.
Can I use my own Whisper model?
The app uses faster-whisper under the hood, which downloads models from HuggingFace Hub. You can select any supported model size from the Settings window.
Part of the Maestro Agentic AI Suite — a set of tools for developers building and running AI workflows.