RednBlue Desktop

Local Files

My Computer

🧠 = model file (click to select)
📁 = folder (right-click for images)

Your model never leaves this machine

Authentication

Don't have a token? Get one

Model Selection

Click to select .pt or .pth file

Test Images

Click to select images folder

Model Type

Tier

Labels file (CSV: filename,class_index) optional

With labels the score is robust accuracy and the grade is not capped. Without labels it is prediction stability (does the output change under attack), and the grade is at most SILVER.

Input size

Prediction stability % = share of tests where the model's output did not change under attack (no labels). Robust accuracy % = share of tests the model got right under attack (with labels). Higher is better.

Target

Click to select .py target file

Function name (optional — auto-detected)

Attack Classes

Freelancer tier: first 3 enabled. Enterprise: full suite.

JB Jailbreak Resistance FREE SPE System Prompt Protection FREE TM Token Manipulation FREE LPI Indirect Injection ENT RAG RAG Pipeline ENT AGT Agent Hijacking ENT BD Backdoor Detection ENT EXT Model Extraction ENT

Skip recon phase (faster, less context)

Submitting LLM results needs RednBlue dashboard v3.1 or later. If the submission fails, save the results locally (output path below) and re-submit later.

Results will be saved to llm_results.json in the working directory.

Cost: 1 Token per test

System Resources

CPU

GPU

RAM

Network: used only to submit results

Data Handling

Model weights stay on your machine
Per-test metrics sent encrypted, only on submit
No images or file paths are uploaded

Compliance

🇪🇺 EU AI Act 🇺🇸 NIST

Local Execution

model.pt

Being tested

images/

0 images

Model never leaves your machine

All processing is local

No network during test

Your Model

Testing...

Attack

Security Assessment in Progress

Progress 0%

Dimensions Completed

Live Resources

Live

CPU 0%

GPU N/A

RAM 0%

Offline Mode

No data being sent

Elapsed Time

00:00

🥇

GOLD

100.0%

Robustness Score

Scenarios

Dimensions

Robustness Radar

Actions

This is a local preview. Submit to the server for a RednBlue certificate (self-reported results, graded by RednBlue, not independently verified).

Detailed Results

Dimension	Tests	Pass Rate	Status