Project Kaizen

The AI That Stays Home

0

Unified Memory

0

Memory Bandwidth

0

Private

0

Active Services

Current Active Stack

max:voice

Active

Qwen3.5-35B-A3B — voice-optimized with fast responses.

Size: ~23GB (Q4_K_M) — 3B active params
Speed: 42.9 tok/s
Role: Primary Voice Assistant

max:deep

Active

Qwen3.5-397B-A17B — Mixture-of-Experts deep reasoning model.

Size: ~189GB (Q3_K) — 17B active of 397B total
Speed: 17.6 tok/s
Role: Primary Reasoning Model

max:think

Active

Qwen3.5-397B-A17B — extended thinking with visible chain-of-thought.

Size: ~189GB (Q3_K) — 17B active of 397B total
Speed: 17.5 tok/s
Role: Deep Analysis with <think> Reasoning

Plus 11 cloud models via Claude, Codex, and Z.AI proxies

Why Local AI?

Kaizen vs Cloud

  • ✓
    Zero Privacy Risk Your data never leaves your Mac Studio. No training on your chats.
  • ✓
    Zero Latency Direct Metal GPU access means instant responses without network lag.
  • ✓
    Zero Censorship Unfiltered, unaligned models that obey YOU, not a corporate policy.

397B

Parameters, 17B Active

512GB Unified Memory runs a 397-billion parameter Mixture-of-Experts model locally — only 17B active per token for datacenter-scale intelligence at conversational speed.

Key Capabilities

🎤

Advanced Voice Mode

Real-time conversation with interruptions, emotion, and sub-400ms latency.

🌐

Web Search Proxy

3-provider fallback (SerpAPI, Brave, Google PSE) with Knowledge Graph, weather, and sports scores.

🧠

Persistent Memory

Mem0 + ChromaDB vector database stores long-term context about you and your preferences.

☁

Cloud AI Proxies

Access Claude 4.6, GPT-5.2, and GLM-5 models alongside local inference via OpenWebUI.

🏠

Home Automation

Deep integration with Hubitat for voice-controlled smart home management.

💡

Intelligent Context

Conditional hardware injection, personal memory recall, and identity enforcement on every request.

Kaizen AI Mobile

Native iOS Experience

Connect to your Mac Studio from anywhere. Kaizen AI v2.3 brings the full power of Max to your pocket — with cloud AI proxy support.

  • 📱 Advanced Voice Mode with live visualization
  • 🧠 Cloud AI Integration — Claude, Codex, Z.AI
  • 🌊 Dynamic Visualizer (Orb, EQ, Waveform, Fluid)
TODAY
Hey Max, what's on my schedule?
mic
end

The Powerhouse

Apple Mac Studio M3 Ultra

  • 512GB Unified Memory
  • 80-Core Metal GPU
  • 32-Core Neural Engine
  • 819 GB/s Memory Bandwidth

Ready for the Future?

Experience the power of unconstrained local AI.

Learn More View Benchmarks