BABILON — The Tower

iOS + macOS · Animal Translate · QUANT VoiceGen · On-Device · Zero Cloud
iOS 18+ macOS 15+ Apple Silicon MLX-QUANT On-Device Zero Cloud

📱 Core Features

Camera: Take pictures, record video — AVFoundation native, full quality, HDR, 4K60.

QUANT VoiceGen: Voice generation powered by MLX-QUANT Metal kernels. Ternary-quantized TTS running on-device. No server. No API key. Your voice, your device.

{-1,0,+1} Filters: Ternary matrix image/video filters — quantize any frame to three values in real time.

MEM8 Wave Memory: On-device, deterministic, zero-network wave interference recall for your media library.

Honesty-Auth: Login via 17-field personality vector, not password. You can't steal someone's mother relationship.

⚙️ Architecture

SwiftUI — Native iOS/macOS interface, MenuBarExtra + WindowGroup

MLX-QUANT — maybe/mergeq Metal GPU kernels for real-time ternary inference

ayeOS — MEMNET protocol daemon on :9876 for ternary matrix serving

entheai — Recursive agent for fan-out video processing

honest-irc — Honesty-auth identity verification

MEM8 — Wave interference media recall, Rust port in Swift

hf-mac — HuggingFace model pulling, Osaurus local inference

🎙️ QUANT VoiceGen Pipeline

1. User speaks → Apple native STT (SFSpeechRecognizer)

2. Text → MLX-QUANT ternary embedding (LINOSV seed, {-1,0,+1})

3. Voice generation → kokoro-tiny TTS (8b-native Rust sidecar) or Apple AVSpeechSynthesizer

4. Output → quantized audio stream layered onto video timeline

5. Export → H.265 + AAC, full quality, local only

8b-is Constellation Stack:
MLX-QUANT · ayeOS · entheai · honest-irc · hf-mac · kernel8 · vaked
License: MIT · CC-BY-4.0 · ALL OSS FREE FOR ALL
Repository: github.com/peterlodri-sec
Status: Scaffolded · Coming Soon · v0.1.0