Local Realtime Voice

Local realtime voice and dictation for OpenClaw using Whisper STT, Kokoro TTS, and Ollama chat.

Install

openclaw plugins install clawhub:@snoutfirst/local-realtime-voice

Local Realtime Voice

An installable ClawHub / npm plugin for OpenClaw that adds a fully local realtime voice assistant and dictation provider.

STT: Whisper via a Whisper-compatible HTTP API (e.g. faster-whisper-server, whisper.cpp server, etc.).
TTS: Kokoro via the /v1/audio/speech endpoint.
Chat: Ollama for local LLM responses.

This plugin was originally proposed as an OpenClaw core extension in PR #96173. The maintainers recommended publishing it as a community plugin instead, so this package is the standalone ClawHub submission. The original PR is left intact and closed.

Install

bash

openclaw plugins install clawhub:@snoutfirst/local-realtime-voice
# or from npm:
openclaw plugins install npm:@snoutfirst/local-realtime-voice

Restart the Gateway after install:

bash

openclaw gateway restart

Configure

Enable the plugin and set the endpoints for your local services in ~/.openclaw/openclaw.json:

json

{
  "plugins": {
    "enabled": true,
    "allow": ["local-realtime-voice"],
    "entries": {
      "local-realtime-voice": {
        "enabled": true,
        "config": {
          "whisperBaseUrl": "http://127.0.0.1:8000",
          "kokoroBaseUrl": "http://127.0.0.1:8880",
          "ollamaBaseUrl": "http://127.0.0.1:11434",
          "chatModel": "llama3.1",
          "kokoroVoice": "af"
        }
      }
    }
  },
  "voice": {
    "realtimeProvider": "local"
  }
}

Options

Option	Default	Description
`whisperBaseUrl`	`http://127.0.0.1:8000`	Base URL of a Whisper-compatible STT server.
`whisperModel`	`""`	Optional model name passed to the STT endpoint.
`kokoroBaseUrl`	`http://127.0.0.1:8880`	Base URL of a Kokoro-compatible TTS server.
`kokoroVoice`	`af`	Voice ID for Kokoro TTS.
`ollamaBaseUrl`	`http://127.0.0.1:11434`	Base URL of the local Ollama API.
`chatModel`	(from config defaults)	Ollama model name for chat responses.
`vadThreshold`	`100`	PCM16 energy threshold for voice activity detection.
`silenceMs`	`1200`	Silence duration that ends a user turn.
`maxTurnMs`	`15000`	Maximum duration of a single user turn.
`partialIntervalMs`	`2000`	Interval for partial dictation transcripts.
`audioChunkMs`	`50`	TTS audio chunk size for streaming playback.

Requirements

OpenClaw Gateway >= 2026.6.9
A running Whisper-compatible STT server
A running Kokoro-compatible TTS server
A running Ollama instance with the model you select

Development

bash

cd local-realtime-voice-clawhub
openclaw plugins install --link .
openclaw gateway restart

License

MIT