What problem does it solve? Building a voice assistant on top of Hermes Agent requires wiring together a fast LLM endpoint, speech-to-text, text-to-speech, and wake-word detection, and Hermes v0.20.x rejects common configuration patterns like a first-class groq provider string. This Skill provides the validated configuration schema, auth commands, and latency tuning needed to get a sub-2-second speech-to-sound assistant running without hitting the documented pitfalls. ## Core Features & Use Cases - Custom Groq endpoint wiring: Registers a custom:groq-ev provider via custom_providers in config.yaml, routing queries through the full Hermes agent loop (persona, memory, tools) rather than a bare LLM call. - Local voice stack: Uses faster-whisper tiny for STT, Edge TTS with a pinned voice, and the sherpa open-vocabulary wake-word engine requiring zero training. - Latency tuning: Documents a full latency budget (target under 2s) with voice-mode keys like silence_duration, barge_in, and streaming TTS that overlaps LLM generation. - Use Case: You want a talking companion AI on your machine that remembers context and uses tools. Follow the setup steps to create a Hermes profile, register the Groq API key, install voice dependencies, and say the wake phrase to get spoken replies. ## Quick Start Ask the agent to set up a Hermes voice assistant profile using a custom Groq endpoint with wake word, local Whisper STT, and Edge TTS following this skill's config schema.
