dsh-voice-input-plugin

Writing long prompts or replies in the DeepSeek Harness Web GUI can feel slow when you only have a keyboard. dsh-voice-input-plugin adds a microphone button to the input toolbar, turning speak-it-see-it-typed into reality — no external recording tool, no server-side recognition service, everything stays in the browser.

Tap the mic and recognition results stream word-by-word into the current conversation draft, appended after any text you already typed. While recording the button turns into a red stop square; on error, hover to see the reason and tap again to retry. Under the hood it uses the Web Speech API in continuous mode, auto-resuming after silence until you stop. Zero dependencies, no build step, no audio ever leaves the browser, and no model tools or prompt segments are registered.

If you work in Chrome or Edge and often dictate longer passages — drafting system prompts, organizing thoughts, typing out spoken questions — this plugin lets you focus on expression instead of typing. It changes nothing else in the workflow: the original send button still fires the message, and uninstalling restores the GUI to its previous state.