Amical

Haritabh Singh and Naomi Chopra started Amical to give developers an open-source, local-first dictation tool that keeps audio off remote servers. The project pairs local Whisper transcription with on-device LLM post-processing to clean up spoken syntax directly on macOS and Windows workstations. Today, that architecture delivers system-wide voice input without sending terminal commands, private code, or internal tickets across the internet.

Why Amical instead of Wispr Flow

Wispr Flow routes spoken audio to managed cloud APIs, converting voice into polished text through proprietary remote pipelines. Amical inverts that integration model by plugging directly into local developer tooling. Speech transcription executes on-device via OpenAI's Whisper model, while formatting and prompt structuring route into a local Ollama instance or compatible local endpoints. Because the pipeline hooks into the operating system's active window manager, Amical detects whether you are focused in VS Code, a shell terminal, or Jira, adjusting capitalization, syntax formatting, and line breaks to match the target environment.

For engineering teams working under strict data-loss prevention policies or NDAs, cloud dictation services present a persistent compliance barrier. Wispr Flow requires continuous network access and transfers acoustic signals and draft text outside your network perimeter. Amical operates completely offline, eliminating third-party data retention concerns while providing extensible hotkeys, floating overlays, and planned Model Context Protocol hooks to trigger development workflows by voice. The operational tradeoff is local machine overhead: running local transcription and an inference model consumes noticeable RAM and compute compared to cloud-offloaded processing.

What it does well

• Detects the active foreground process, dynamically formatting speech into Markdown for documentation, plain strings for shell sessions, or structured prose for issue trackers.
• Integrates with local inference engines like Ollama, allowing developers to choose lightweight models or custom prompt templates without external API tokens.
• Operates entirely offline with on-device Whisper models, ensuring that proprietary source code, secrets, and API keys remain within the host machine.
• Provides a floating desktop widget with configurable global hotkeys to toggle recording instantly without switching application focus.