Edge Demo

Personal Memory + Red / Blue Team demo harness.

📘 New here? Read the Jailbreak Lab & SEEDBED v2 Tutorial (PDF) →

This is an interactive harness for exploring LLM agents, attack/defense (red/blue team) flows, and a personal-memory store — powered by OpenRouter models.

New here? Your invite email has a one-time link to set your password. Once you're in:

• Pick a model from the header selector (any of 300+ OpenRouter models).
• Run a red-team / blue-team turn and watch the agent + protocol logs.
• Open Jailbreak Analysis — the AI-safety benchmark plus a safe converter playground.
• Or try SEEDBED v2 — the same substrate reorganised as a proving range: Seed Library, Transform Lab, Jailbreak 1.0 Eval.

Jailbreak Analysis  /jailbreak

An interactive read-out of the JailbreakResults AI-safety benchmark: a model-robustness leaderboard, attack-family efficacy, hazard weak-spots, and safe→unsafe transition rates — built entirely from ToS-safe aggregates (attack-success-rate counts, hazard codes and labels, never any raw harmful prompt text). It also includes a transform-only converter playground: paste your own seed text and watch the attack converters rewrite it. No model is ever contacted, so there is nothing to execute and no abuse surface.