A high-performance, real-time voice interface for Openclaw agents featuring sub-2 second latency and multi-provider STT/TTS support.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install voice-assistant
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install voice-assistant using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Voice Assistant?
The voice assistant is a sophisticated communication layer designed for Openclaw Skills that enables developers to interact with their AI agents using natural speech. By bridging the gap between browser-based audio capture and agent logic, this skill allows for a fluid, hands-free experience. It utilizes WebSockets to maintain a persistent connection, ensuring that audio data is streamed bi-directionally with minimal overhead.
This skill is not just a simple wrapper; it integrates deeply with the OpenClaw gateway, maintaining full access to the agent's context, tools, and memory. Whether you are using Deepgram for ultra-fast processing or ElevenLabs for high-fidelity vocal synthesis, this tool optimizes the entire pipeline to deliver a conversational experience that feels responsive and human-like within the ecosystem of Openclaw Skills.
Voice Assistant Use Cases
- Hands-free coding and environment configuration by issuing voice commands to your agent.
- Real-time collaborative brainstorming where you can speak ideas and hear the agent's analysis instantly.
- Improving accessibility for developers who prefer voice-to-text workflows over traditional keyboard input.
- Deploying low-latency voice interfaces for custom AI tools built on Openclaw Skills.
How Voice Assistant Works
- The browser captures raw mic audio via the Web Audio API and streams it to the server using a WebSocket.
- The server forwards the stream to a selected Speech-to-Text (STT) provider such as Deepgram or ElevenLabs for real-time transcription.
- Upon detecting the end of an utterance, the transcribed text is sent to the OpenClaw gateway's OpenAI-compatible endpoint.
- The gateway processes the request and streams the response tokens back to the voice server via Server-Sent Events (SSE).
- As tokens arrive, they are aggregated into sentence chunks and immediately dispatched to a Text-to-Speech (TTS) provider.
- The resulting audio chunks are streamed back to the browser and played through a jitter-buffered audio pipeline for smooth playback.
- Integrated interruption handling ensures that if the user starts speaking again, the current audio playback is cancelled to begin a new turn.
Voice Assistant Setup
To get started with this voice interface for Openclaw Skills, ensure you have the uv package manager installed.
# Install the uv dependency manager
brew install uv
# Navigate to the skill directory and prepare environment
cd {baseDir}
cp .env.example .env
# Edit .env to include your API keys and gateway URL
# OPENCLAW_GATEWAY_URL=http://localhost:4141/v1
# VOICE_STT_PROVIDER=deepgram
# Launch the voice server
uv run scripts/server.py
Open your browser to http://localhost:7860 to begin interacting.
Voice Assistant Data Schema & Taxonomy
The skill manages its operation through environment variables and structured WebSocket messages. Below is the primary configuration schema:
| Variable | Description | Default/Options |
|---|---|---|
| OPENCLAW_GATEWAY_URL | The API endpoint for your agent | http://localhost:4141/v1 |
| VOICE_STT_PROVIDER | Provider for Speech-to-Text | deepgram, elevenlabs |
| VOICE_TTS_PROVIDER | Provider for Text-to-Speech | deepgram, elevenlabs |
| VOICE_VAD_SILENCE_MS | Milliseconds to wait for silence before processing | 400 |
| VOICE_SAMPLE_RATE | Audio sampling frequency in Hz | 16000 |
| VOICE_TTS_VOICE | Voice ID for the chosen TTS provider | rachel, aura-2-theia-en |
Voice Assistant Advanced Features
- Barge-in support: Allows users to interrupt the agent at any time, immediately halting TTS output to listen to new input.
- Multi-provider flexibility: Mix and match different providers for STT and TTS to balance between cost, speed, and voice quality.
- Performance tuning: Adjust Voice Activity Detection (VAD) thresholds and sample rates to optimize Openclaw Skills for specific network conditions.
- Latency Budgeting: Designed to maintain a total time-to-first-audio of under 2.5 seconds, even with complex LLM reasoning.
SKILL.md
Loading
Related Openclaw Skills
YouTube Summary
A specialized tool for extracting and summarizing YouTube transcripts to save time on video content consumption.
Typefully Social Media Manager
A powerful automation bridge for Typefully to manage social media drafts and multi-platform scheduling directly via the CLI.
Typefully Social Media Manager
A powerful CLI-driven tool for managing Typefully drafts and scheduling across multiple social platforms.
LLM Speedtest
A high-performance utility to measure and compare Time to First Token (TTFT) latency across major LLM providers in parallel.
People Memories Skill
A lightweight personal CRM and context-retention skill for AI agents to store and recall specific details about individuals.
Copilot Money Finance Analyzer
Query and analyze local Copilot Money transaction and budget data using SQL and shell utilities.
Featured*
