dsh-voice-input-npm

Typing often lags behind thinking, and prolonged keyboard work fatigues your fingers. If you want to dictate prompts straight into the DeepSeek Harness composer without spinning up a single speech-recognition backend or requesting an API key, this plugin is built for that exact moment. Once installed, a compact microphone button appears in the input toolbar. Under the hood it leans entirely on the browser-snative Web Speech API - no server, no key, no recurring bill. Open Chrome and you are ready to go.

Click the button and it turns red with a gentle pulse animation, signalling that it is listening. As you speak, words appear in the draft area in real time. The transcription is appended to whatever text you already have, so a sentence you started by hand is never wiped away. Click the button again (the icon changes to a stop square) to end recording, review the text, and hit Enter to send as usual. The button colour follows DSH theme variables so it blends seamlessly into both light and dark modes. The default recognition language is Simplified Chinese; a one-line change to rec.lang in lib/client.js switches it to English.

The ideal user is straightforward: you work in DSH daily with Chrome or Edge, you dictate long prompts often, and you would rather not operate a WebSocket speech service just for this one convenience. The first click triggers a microphone permission prompt - simply tap Allow. Firefox does not support the Web Speech API, so a Chromium-based browser is recommended. If fully offline recognition and stronger privacy guarantees are on your list, keep an eye on the roadmap for a planned Vosk engine integration.