What it does

Amical is an Electron-based dictation app that transcribes speech with OpenAI’s Whisper entirely on your own hardware. It detects which application is active—email client, IDE, chat app—and adjusts formatting so your dictated text lands in the right shape for the context. A floating widget handles start/stop without pulling you out of flow.

The interesting bit

The “context-aware” angle is the hook: instead of dumping raw transcription and making you clean it up, Amical tries to infer intent from the foreground window. It’s a small UX bet that voice input fails more on friction than on accuracy. The planned MCP integration suggests they want to push this further toward voice-controlled automation.

Key highlights

• Runs offline via local Whisper + Ollama; no cloud audio shipping
• One-click model setup inside the app
• Custom hotkeys and voice macros for extensibility
• Cross-platform: macOS, Windows, Android, iOS beta
• Smart voice notes (summaries, tasks) marked as in-progress

Caveats

• Several headline features are still planned or in progress: MCP integration, real-time meeting transcription with system audio, and smart note structuring
• The “3x faster” claim in the repo description is unsupported by any benchmark data in the README

Verdict Worth a look if you dictate regularly and would rather not send audio to a remote API. Skip it if you need mature meeting transcription or deep voice automation today; those pieces aren’t baked yet.