dsh-omi-voice

If you're tired of DeepSeek Harness's built-in voice reading that sounds mechanical and dated, and you don't want to manually copy every reply just to hear it, dsh-omi-voice offers a smoother way. It adds a small 🔊 button next to each AI reply, letting you hear the answer read aloud in Doubao's natural Chinese voice with a single click.

Rather than being a simple playback widget, the plugin acts as a lightweight remote control: play, pause, resume from where you left off, and interrupt by reading a new message are all handled by the local Omi engine running on your Mac. It only reads final answers, automatically skipping tool execution logs and thinking traces, and filters out code fences, tables, and pure ASCII graphics before making a request. If a reply has nothing worth reading, the button is disabled with a clear hint instead of wasting a request.

The plugin takes privacy and cost seriously with a BYOK model: your Doubao API key stays only in the macOS Keychain, while the plugin itself holds no keys and communicates exclusively over local 127.0.0.1. Speech synthesis is billed per character, but manual triggering, 3-second deduplication, and an in-memory LRU cache help avoid repeated charges for the same content.

If you use DeepSeek Harness on macOS and want AI replies to sound more like a real person than a robotic announcer, and you're comfortable bringing your own Doubao API key and paying per use, this plugin is a great fit. Note that it currently only supports macOS on Apple Silicon.