ZDX Mobile AI
live on S25 UltraFlutter · Android · On-Device AI
ZDX Mobile AI
Your phone. Your AI. No cloud. No compromise.
Built in Flutter, running llama.cpp natively via JNI — no cloud, no API calls. A dedicated 0.5B scorer model blended with instant heuristics ranks every memory 1–20, so your AI keeps what matters and discards the noise across sessions. Vulkan GPU pipeline targets Samsung Galaxy S25. Runs on Android 8+ down to mid-range hardware. Pull SIM. Kill WiFi. It works.
Live demo — Vulkan GPU acceleration
What Makes It Different
On-Device AI Done Right
Vulkan GPU
Upgraded Vulkan pipeline optimized for Samsung Galaxy S25. Also runs faster on older Android phones — not just flagships.
Gravity Wells
Adaptive routing across Qwen2.5 and Llama 2 based on task complexity — fast answers or deep reasoning, automatically.
On-Device Tools
SMS, calls, contacts, calendar, GPS, camera, web search and more — all executed locally via native Android.
Zero Telemetry
No cloud. No API. No training on your data.
Proprietary Technology
Memory Management Engine
Every exchange is scored 1–20 by a dedicated on-device 0.5B model, blended with instant heuristics — tool results score high, filler gets pruned. The KV cache eviction engine uses age-weighted scoring so important context survives across sessions. On low-RAM devices the pruner gets more aggressive, protecting critical context while dropping noise. Your AI remembers what you need, not everything you said.
1–20 Scoring
Tool results, instructions, and facts rank highest. Greetings and filler are pruned.
Session Continuity
High-scoring chunks survive KV cache eviction and persist across conversations.
RAM-Aware
Eviction aggressiveness scales to available memory — works on mid-range hardware.
Ready for AI that belongs to you?
First to know when we launch on Indiegogo