Phantom Orbital: A floating Windows orb that lets you talk to your AI assistant out loud.
Phantom Orbital is a small, floating desktop companion for Windows 11 that gives a voice and desktop presence to an AI agent harness like Hermes, paired with either a cloud LLM (such as OpenAI) or a local LLM runner like Ollama or LM Studio. You can type to it, hold a shortcut key to talk, or wake it up hands-free with a wake phrase.
How the Orb Looks in Action
Click through the thumbnails to see how Phantom Orbital changes visually as it listens, thinks, and speaks in both dark and light themes.
Listening Mode: Activated by holding Ctrl+Shift+Space, saying your wake phrase ("Hey Megan"), or automatically for 5 seconds after the assistant finishes speaking so you can reply naturally.
A Friendly Face and Voice for Your AI Agent
Talking to an AI agent on your computer shouldn't require digging through terminal windows or streaming raw microphone audio to the cloud.
If you run an AI agent harness on your computer (like Hermes), interacting with it usually means switching windows and typing into a console or chat tab. I wanted something that felt like a true desktop companion, always within reach when you want to ask a quick question or kick off a task out loud.
At the same time, I didn't want an always-listening cloud microphone on my desk. So I designed Phantom Orbital to handle all of the voice work (listening for the wake word, turning speech into text, and reading replies out loud) right on your own PC, and then send only the transcribed text to your agent harness. From there, you have full flexibility over the LLM brain: for instance, my daily setup uses Hermes as the agent harness and OpenAI as the brain, though you can just as easily point your harness at a local LLM interface like Ollama or LM Studio if you want a 100% local stack.
Local Voice, Flexible Brain
Your microphone audio never leaves your PC, speech recognition and voice synthesis run locally. Pair your agent harness with a cloud model like OpenAI, or go 100% local through Ollama or LM Studio.
Click-Only Approvals
When your AI agent asks for permission to run a tool or command, Orbital pops up an Approve once / Deny card on screen and never approves actions by voice.
Mess-Free Setup
The setup scripts install everything the voice engine needs inside the project's own folder, without messing with your computer's existing settings.
How Conversations & Personas Work
Hands-free follow-ups, optional voice matching, and customizable assistant personalities.
1. A Natural Conversation Flow
Start Talking
Hold Ctrl+Shift+Space to talk, say the wake phrase ("Hey Megan"), or just type a quick message.
Local Speech-to-Text
Your spoken words are turned into text on your PC (with an optional filter that only responds to your own voice).
Send to Agent Harness
The text goes to your agent harness (such as Hermes), backed by your choice of cloud LLM (e.g., OpenAI) or local LLM interface (Ollama, LM Studio).
Speak & Follow Up
The orb reads the answer out loud, then listens for 5 seconds so you can ask a follow-up question without repeating the wake phrase.
2. Swappable Persona Packs
You're not locked into a single assistant style. Each persona is just a simple folder with two files: a persona.json file (which sets the assistant's name, wake phrase, speaking voice, speed, and quick "thinking" phrases) and a plain-text personality.md file describing how the assistant should talk and behave.