openclaw - 💡(How to fix) Fix [Feature]: Support Gemini Live / realtime audio providers for Discord voice channels [1 participants]

Summary

Add support for using Gemini Live / realtime audio providers as the backend for Discord voice channel conversations.

Today, Discord voice appears to use a text-mediated pipeline:

Discord voice audio -> STT -> text LLM response via channels.discord.voice.model -> TTS -> Discord voice playback

The new channels.discord.voice.model config is useful for choosing the text response model, but it does not appear to select an end-to-end realtime audio backend like Gemini Live.

Motivation

OpenClaw now has realtime voice support in adjacent surfaces:

• Voice Call / realtime mode
• Google Meet audio bridges
• Talk / realtime voice loops
• Gemini Live as a realtime voice provider for backend Voice Call and Google Meet audio bridges

It would be useful for Discord voice channels to be able to use the same realtime audio provider path, rather than always going through separate STT -> text LLM -> TTS stages.

This could reduce latency and make Discord voice channel interaction feel closer to native realtime voice systems.

Proposed behavior

Provide a Discord voice configuration path that can select a realtime audio provider/session, for example Gemini Live, instead of only selecting the text LLM used after STT.

The exact config shape may differ, but conceptually this would be separate from channels.discord.voice.model, which should remain the text LLM override for the current pipeline.

Potential shape:

{
  "channels": {
    "discord": {
      "voice": {
        "mode": "realtime",
        "realtime": {
          "provider": "google",
          "model": "gemini-2.5-flash-native-audio-preview"
        }
      }
    }
  }
}

Current understanding

From the current schema, channels.discord.voice includes:

• enabled
• model
• autoJoin
• DAVE encryption settings
• tts

The model field is documented as an optional LLM override for Discord voice channel responses, e.g. openai/gpt-5.4-mini, while STT and TTS remain on their existing media settings.

That suggests Gemini Live-style realtime audio is not currently configurable for Discord voice channels.

Related issues

Related, but not exact duplicates:

• #60093: voice-call plugin support for Google Gemini Live as an end-to-end audio provider
• #45561: native Gemini Live API support
• #7200: real-time voice conversation support
• #71262: voice-call realtime mode should expose gateway agent tools

Those issues cover Gemini Live / realtime voice generally, especially Voice Call. This request is specifically for Discord voice channels.

Open questions

• Should Discord voice realtime mode reuse the same realtime voice bridge/session runtime as Voice Call and Google Meet?
• How should Discord voice realtime sessions preserve existing Discord channel context, permissions, and agent routing?
• Should realtime mode still emit transcripts for logs, moderation, memory, and thread context when the provider is audio-native?
• How should tool calls and agent consults be bridged so Discord realtime voice does not lose the normal OpenClaw agent capabilities?