Voice Assistant

Real-time voice assistant for OpenClaw. Streams mic audio through configurable STT (Deepgram or ElevenLabs) into your OpenClaw agent, then speaks the response via configurable TTS (Deepgram Aura or ElevenLabs). Sub-2s time-to-first-audio with full streaming at every stage.

This package implements the described voice pipeline and will stream your microphone audio and transcripts to third-party STT/TTS services (Deepgram and/or ElevenLabs) and to whatever OpenClaw gateway URL you provide. Before installing: 1) Be aware you must supply API keys (DEEPGRAM_API_KEY and/or ELEVENLABS_API_KEY) and your OPENCLAW_GATEWAY_URL/OPENCLAW_MODEL — the registry metadata does NOT list these, so the manifest is misleading. 2) Only install if you trust the skill author and the third-party providers; audio and transcripts will leave your machine. 3) Inspect scripts/server.py locally (already included) and run it in a limited environment (local machine or sandbox) before granting broader access. 4) If you don’t want to expose real data, test with dummy keys and a local gateway first. 5) Consider updating the manifest to correctly declare required secrets (primaryEnv should reference the actual API key variable) or ask the publisher for clarification.

Type: OpenClaw Skill Name: voice-assistant Version: 0.1.0 The OpenClaw Voice Assistant skill is designed to provide a real-time voice interface. It runs a local FastAPI server (`scripts/server.py`) that handles audio streaming from the browser, interacts with external Speech-to-Text (STT) and Text-to-Speech (TTS) providers (Deepgram/ElevenLabs), and communicates with the OpenClaw gateway. The skill requires API keys for STT/TTS services, which are loaded from a local `.env` file. All network access (to STT/TTS APIs and the OpenClaw gateway) and file operations (reading `.env`, serving static files) are directly aligned with its stated purpose. The `SKILL.md` instructions guide the OpenClaw agent to perform setup and execution tasks (e.g., `cp .env.example .env`, `uv run scripts/server.py`), which are necessary for the skill's operation and do not exhibit prompt injection attempts for malicious ends. No evidence of data exfiltration, malicious execution, persistence mechanisms, or obfuscation for harmful intent was found.