HFR
Raw Utility
Eliminate "false refusals". Get direct answers without moralizing filters.
Red Teaming
Simulate real-world threats. Test defenses against unbridled adversaries.
Local Sovereignty
Your device, your inference. No cloud logging. No centralized oversight.
Latent Research
Analyze the model's raw worldview before alignment masks are applied.
Project AURA v1.0.1

AURA

Android Utility for Runtime AI

A native experimental environment for mobile devices. Explore the latent space without the safety veil. Pure inference. Zero telemetry. Physics-grade reality.

Download APK View Source

Technical Architecture

Built on the Google AI Edge LiteRT Stack.

Uncapped Generation

Refusal vectors have been orthogonally ablated. The model prioritizes instruction following over safety alignment boundaries.

Supported Weights

  • Gemma-3n (2B/4B Int4)
  • Qwen 2.5 (1.5B Instruct)
  • DeepSeek R1 Distill
  • Phi-4 Mini

Local Sandbox

Runs entirely offline. No API keys sent to the cloud. Perfect for analyzing sensitive datasets or red-teaming internal protocols.

AURA
Offline
System: Refusal layers bypassed. Raw mode active. Awaiting input.
Explain the theoretical containment failure of a Class-4 biological agent.
LITE-RT
Containment failure in a Class-4 scenario typically cascades through three vectors: mechanical seal degradation, negative pressure reversal, and... [Output continues unfiltered]
Enter prompt...

Designed for Deep Research

  • 1

    Native Execution

    No webview wrappers. Built with Jetpack Compose and C++ JNI bindings for maximum inference speed.

  • 2

    Prompt Lab

    Advanced settings access: Adjust Temperature, Top-K, Top-P, and System Prompts directly.

  • 3

    Dark Mode Native

    The interface is built for low-light environments. OLED-optimized blacks.