Ghost Vessel

What Ghost Vessel does

Ghost Vessel is a monitor-resident, video-call-style avatar window that fronts a local LLM agent instead of a Telegram or messenger chat. The agent's reply is split into three planes: action (emotion beats played as pre-rendered clips), dialogue (spoken through local TTS) and data (code and files shown as chat cards, never read aloud). Because the avatar plays pre-rendered clips, the GPU stays free at runtime.

Typing a slash opens the agent's live command menu, the same one seen in Telegram. For Hermes the README cites 52 commands, such as /model, /new, /goal, /status and /compress, and commands pass straight to the agent. Adapters exist for Hermes (a relay connector contract with a WebSocket server) and OpenClaw (a gateway WebSocket client), plus a demo responder that needs no setup. A persistent mood tracker sinks when you scold the avatar and recovers when you praise it. Voice can be local (Qwen3-TTS with voice cloning, MeloTTS, Piper) or cloud (Edge, ElevenLabs, any OpenAI-compatible service), with voice input through VAD and local STT.

Key features

• Video-call-style avatar window with emotion beats and a mood that persists across restarts
• Three-plane output that separates action, spoken dialogue and data cards
• Serves the agent's live slash-command menu inside the avatar window
• Adapters for Hermes and OpenClaw, plus a demo responder that needs no agent
• Swappable local or cloud text-to-speech, with voice input through VAD and local STT
• Pre-rendered clips, so no live inference or GPU is needed at runtime

When to use it

• Replacing a Telegram chat window with a face for a personal Hermes agent
• Running Hermes slash commands such as /model or /status from an avatar window
• Adding a custom avatar by dropping its folder into the presets directory

Who it is for: People who run a personal local agent such as Hermes or OpenClaw and want a talking avatar front end instead of a messenger chat.

How it fits with Hermes Agent

It supports Hermes Agent through a relay connector and WebSocket server adapter, alongside an OpenClaw adapter, and passes Hermes slash commands straight through.

How to install Ghost Vessel

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

echo gv-starter > presets/active.txt
python bridge/setup_connector.py

Requirements: Python 3.11 on Windows; an agent is optional because demo mode works without one

Note: It is open-core: the repository ships the engine and a free starter avatar, while the Windows package (from $5) and the photoreal preset (from $1) are paid, pay-what-you-want downloads.

FAQ