The core design philosophy of DeepSeek Harness (DSH) is “everything is a plugin.” When building agent applications, voice interaction is a common requirement, typically involving the integration of speech-to-text (ASR) and text-to-speech (TTS). The dsh-voice-mimo plugin introduced below aims to provide DSH with a voice interaction solution based on Xiaomi MiMo technology, covering browser-side voice operations and Agent-side voice tool invocation capabilities.

Plugin Positioning

The plugin was originally named ch1bug/dsh-voice-mimo and is a voice UI and toolset based on Xiaomi MiMo. The repository is currently marked as Deprecated, and its development work has been fully merged into the dsh-mimo-agent-tools repository as part of a single-repository decision.

Core Features

The plugin provides the following capabilities:

  1. Browser Voice UI: Supports browser-native voice input (microphone) and read-aloud (TTS) features, and can play voice clips or cards.
  2. Agent Tools: Provides tools such as voice_transcribe, voice_understand, and voice_speak, allowing voice processing capabilities to be called directly within conversation flows.
  3. MiMo Client: Wraps MiMo TTS and ASR clients and provides a lightweight ctx export.
  4. Configuration and Storage: Supports settings in the plugin tab and provides audio storage.

Installation and Enablement

Before enabling the plugin, installation must be completed. According to the official documentation, the installation command is as follows:

dsh plugin --profile web add github:ch1bug/dsh-voice-mimo

After installation, configure XIAOMI_API_KEY in DSH credentials, then open Settings → Plugins to configure voice-related options.

Typical Usage

After the steps above, the plugin will load automatically.

  1. Browser-Side Interaction: Users can use the microphone to input voice, and the system uses the voice_transcribe tool to convert speech to text.
  2. Agent Invocation: The Agent can call voice_understand to process voice data, or use the voice_speak tool to synthesize generated text into speech and play it back.
  3. Migration Note: The current browser voice UI and Agent tool features have been migrated to the ./client (UI) and ./mimo (tools) directories of the dsh-mimo-agent-tools repository.

Notes

  1. Status Change: The plugin has been archived and no longer receives new updates; all work has been absorbed into dsh-mimo-agent-tools.
  2. License: MIT License, derived from zhuiyueya/dsh-voice and Anionex/dsh-vision-toolkit.
  3. Runtime Dependency: XIAOMI_API_KEY must be configured in DSH credentials before running.
  4. Model Version: Upstream mimo-v2.5 has been retired; V2.6 is currently used by default.

Summary

The dsh-voice-mimo plugin provides a complete voice capability loop from browser UI to Agent tools. Although the current repository has been archived, its core features have been seamlessly migrated to dsh-mimo-agent-tools, where developers can refer to the relevant directories for complete documentation and the latest maintenance.