Status: Active · ← All projects
Jarvis runs entirely on local hardware: the qwen3:14b model on an RTX 4070 SUPER through Ollama, with tools that can read and act on the network, Home Assistant and Docker.
In place
- Fully local voice stack: speech recognition, a British-voiced text-to-speech and a “Hey Jarvis” wake word
- Open WebUI with UniFi, Home Assistant and Docker tools
- A self-improvement loop where every proposed change waits in an approval queue
- Feedback and lessons that become retrievable memory only after approval
Next
- Add a microphone satellite to try real speech
- A confirmation layer and an action log for everything Jarvis does
- Event-driven alerts: diagnose first, then tell me
Last updated 2026-10-03.