CIRCUIT & CO — PROJECT 001 — LOCAL AI SYSTEM

JARVIS

Local AI Operating System

Not an app. Not a chatbot. A presence. A fully local AI assistant that wakes when you call it, sees through your camera, speaks in a cinematic voice, and runs entirely on your own hardware — no cloud, no subscription, no compromise.

● OPERATIONAL GPU LOCAL PRIVACY FIRST ACTIVE DEVELOPMENT
Core Status
ONLINE
Running on local GPU hardware
Voice Engine
KOKORO
Premium British TTS — no cloud
Vision Module
LLaVA
Real-time camera perception
Wake Detection
ALWAYS ON
openWakeWord — passive listening
Live Session

What It Looks Like

jarvis — local session — GTX 1080
[SYSTEM] Initializing JARVIS core... [SYSTEM] Loading qwen2.5 via Ollama... GPU allocated: 6.8GB VRAM [VISION] LLaVA vision model active — camera feed online [AUDIO ] faster-whisper loaded — speech recognition ready [VOICE ] Kokoro TTS initialized — voice profile: British Premium [WAKE ] openWakeWord listening for: "Hey Jarvis" ───────────────────────────────────────────── [WAKE ] Wake word detected. USER > "Jarvis, what do you see right now?" [VISION] Processing camera frame... JARVIS > "Good morning, Mr. Johnson. I can see you're at your desk. It's currently 9:14 AM. Your workspace looks clear — ready to begin when you are, sir." ───────────────────────────────────────────── USER > "What's on my agenda today?" JARVIS > "Checking your schedule now, Mr. Johnson..."
Architecture

The Stack

Agent Layer
OpenJarvis + Ollama
OpenJarvis handles the agent framework and persona logic. Ollama runs qwen2.5 locally on the GTX 1080 — no API calls, no latency spikes, no data leaving the machine.
OpenJarvis Ollama qwen2.5 GTX 1080
Voice Layer
Kokoro TTS + RVC
Kokoro provides premium neural text-to-speech in a natural British accent. A custom RVC voice cloning pipeline was built from scratch to train a signature Jarvis voice on clean audio.
Kokoro TTS RVC Voice Cloning Python
Perception Layer
LLaVA + faster-whisper
LLaVA gives Jarvis real-time vision — it can see through a camera and describe what it observes. faster-whisper handles speech recognition with low latency on local hardware.
LLaVA faster-whisper Computer Vision ASR
Activation Layer
openWakeWord
Always-on passive listening via openWakeWord. Jarvis stays silent until you call it — then it wakes instantly. No cloud processing for wake detection. Fully local, fully private.
openWakeWord Always-on Audio Local Processing
Roadmap

What's Coming

The current build is a working proof of concept — a fully functional local AI assistant that most people have never seen operate this well in person.

The final product vision is a premium personal AI operating system. Not an app you open. A presence that lives in your home — monitoring your schedule, reading your email, controlling your smart home, and getting smarter about you every single day.

All of it running on your hardware, with your data. No monthly subscription for the core experience. No sending your conversations to someone else's server. Just a genuinely intelligent assistant that's entirely yours.

The market gap is real — cloud-dependent competitors are already selling for $78/month. What this builds toward is technically superior in every meaningful way.

🎙️
Wake Word + Voice Conversation
Always listening, responds naturally, maintains full session context
LIVE
👁️
Real-Time Camera Vision
LLaVA perception — sees and describes the physical environment
LIVE
🔊
Custom Voice Clone
RVC-trained signature voice — cinematic, distinctive, unmistakable
LIVE
📅
Calendar & Email Integration
Daily spoken briefings, schedule awareness, email management
SOON
🏠
Smart Home Control
Home Assistant integration — lights, locks, thermostat, cameras
SOON
🧠
Persistent Memory
Learns preferences, routines, and context over time
SOON

See More Work

Interested in what else Circuit & Co is building? View the full portfolio or get in touch directly.

← Back to Portfolio
Demo video coming soon — the voice has to be heard to be believed.