LevelUp generates fresh Docker-sandboxed challenges every night and reconstructs published breach reports as multi-stage scenarios within days. Red, Blue, or Purple stream — every solve calibrated by a 12-axis skill vector.
Nine coordinated agents generate, validate, and calibrate every challenge before you see it — or take a breach report as input and reconstruct it as training within days.
LLM agent drafts the challenge brief — vulnerable app, malicious binary, or incident packet.
Deterministic linter catches broken Dockerfiles, missing flags, and obvious generator slop before build.
Image builds, services come up healthy, auto-solver reaches the flag end-to-end.
Par time measured, difficulty scored against target ELO band, gap closed or sent back.
Stage failure? Repair agent pulls institutional memory and patches the draft instead of regenerating.
Hardened container image ships to the library with stream labels and difficulty badge.
Type-aware attack agent proves the flag is reachable via HTTP, crypto solver, binary exploit, or JS contract call.
Skill-vector-aware hint agent generates 3 progressive hints tied to your weakest axis.
Nightly worker runs five loops: recalibrate, mutate, evolve prompts, fill difficulty gaps, retire miscalibrated.
Or paste a breach report URL — REACTOR reconstructs it as a multi-stage scenario. See the full pipeline →
One platform. Same ELO. Same skill vector. Offence, defence, or both.
Find the bug, chain the primitives, capture the flag. Real apps, real vulns, real exploit chains.
Triage SIEM alerts, correlate evidence, name the threat. Live captures, real logs, real attackers.
ELO-matched across the full offensive / defensive spectrum. The hardest tier on the platform.
REACTOR's evolution worker runs five loops at 00:01 UTC. Challenges that everyone solves too fast get mutated. Prompts that produce boring briefs get retired. Gaps in the catalogue get filled from real solve-rate data. Challenges that drift out of their calibrated band get demoted or retired.
Every challenge generated by REACTOR. Mutates when solve rates drift too high. No writeups, no pre-indexed solutions, no LLM shortcuts.
Paste a published breach report. REACTOR reconstructs it as a multi-stage scenario within days — fresh training on incidents that happened last week.
Your 12-axis skill vector updates on every solve. The matchmaker queues the next challenge in your growth zone — not where a linear path says you should be.
Dedicated container per session with the right toolkit for the category. Terminal access over WebSocket — no simulations, no multiple choice.
Every challenge has an AI solver par time. Beat it and you earn an Eagle: 2× coins and a mark on your par ratio. Compete against the machine.
Every keystroke. Every retry. Every hint. Captured, classified, and graded. AI-vs-human detection ships as a first-class signal.
Every challenge has an AI par time — the moment our solver captures the flag. Beat that par, and you've got an Eagle. Here's who's running the platform this week.
Free for practitioners. Enterprise pricing transparent on /pricing.