Talk to a Hermes agent by voice and hear its reply streamed back sentence by sentence.
Watch live tool calls and approve or deny dangerous agent actions from the HUD.
Offload speech-to-text transcription to an NVIDIA GPU elsewhere on your LAN for speed.
Summon media panels like videos or images that the agent displays on the HUD.
| eadmin2/jarvis_ai | greatvishal27-rc/ai-resume_analyzer | jason904/ui-skill-lab | |
|---|---|---|---|
| Stars | 56 | 56 | 56 |
| Language | Python | Python | Python |
| Setup difficulty | hard | moderate | moderate |
| Complexity | 4/5 | 2/5 | 3/5 |
| Audience | developer | vibe coder | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires a running Hermes Agent installation, Python 3.11+, and an ElevenLabs API key, designed for LAN-only use.
This repository is a self-hosted voice assistant and browser-based control panel styled after the J.A.R.V.I.S. AI from Iron Man. It runs on top of Hermes Agent, an open-source autonomous agent made by NousResearch that can read and write files, run terminal commands, search the web, and remember conversations across sessions. The project wraps that agent in a voice interface with a glowing circular HUD that runs in any browser on your local network. The setup involves a Python server that handles two things: converting your speech to text locally using the Whisper model (no cloud needed for that part), and streaming the agent's text replies back as speech through ElevenLabs. When you click the ring and speak, your words appear on screen in real time, get sent to Hermes Agent, and the response plays back sentence by sentence as it is generated. A typical back-and-forth takes three to five seconds. The HUD around the ring is a full control panel. It shows live tool calls as the agent works, lets you stop a running agent action mid-stream, and pauses to ask for your approval before the agent runs anything considered dangerous. If the agent is speaking and you want to interrupt, clicking the ring again cuts it off and the agent tracks exactly what you heard. The HUD also embeds Hermes' built-in kanban board and session browser as interactive pop-up panels, and can summon media panels (videos, images) that animate in from a z-depth effect. A Machines panel shows CPU and GPU stats for your host and any connected worker machines. Speech-to-text runs locally on CPU by default. An optional sidecar lets you offload transcription to any NVIDIA GPU on your LAN for much faster results, with automatic fallback to local processing. The system is designed to stay on your LAN only. The Hermes API key never reaches the browser, all endpoints require a token you enter once per device, and secret-looking strings in your speech are redacted before anything reaches the cloud TTS service. The project is MIT licensed.
A self-hosted, Iron Man style voice interface and HUD that lets you talk to the Hermes autonomous agent in your browser.
Mainly Python. The stack also includes Python, FastAPI, Whisper.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.