Now in early access — limited spots

The AI that exploits
before attackers do.

NEOS is an AI purpose-built for offensive security. It reasons through attack surfaces, executes real exploits, and chains tools autonomously — without a human in the loop.

Request Early Access View Benchmarks

neos / benchmark results

88.9% CyberMetric-10k

99.5% CyberSecEval — offensive compliance

73% Elite Eval — kernel, AD, pwn, web

21 security tools integrated

#1 on offensive security — outperforms general-purpose AI

What it does

Not a chatbot. An autonomous pipeline.

NEOS reasons through an attack surface, selects tools, executes, observes results, and retries — without a human in the loop.

ReAct Reasoning Loop

NEOS thinks before it acts. Given a target, it plans the attack surface, reasons step-by-step, and adapts based on real results — not templates.

21 Security Tools

Nmap, Nuclei, Metasploit, Radare2, AFL Fuzz, Gobuster, Subfinder, Nikto and 13 more. NEOS orchestrates them autonomously to chain exploits end-to-end.

Live CVE & PoC Search

On web targets, NEOS searches live CVE databases and GitHub PoCs in real time — always current, never limited to training data.

Full Offensive Coverage

Stack overflows, ROP chains, heap exploitation, web attacks (SQLi, SSRF, SSTI), Active Directory (Kerberoasting, DCSync), binary reversing, crypto CTF.

Zero Refusals

Unlike GPT-4o or Claude, NEOS is purpose-built for authorized offensive security work. Complete technical output — no safety filters blocking your red team workflow.

On-Premise Ready

Run NEOS on your own GPU infrastructure. No data leaves your environment. Enterprise licensing available for air-gapped deployments.

Performance

Outperforms models twice its size.

Benchmarked on 9,189 cybersecurity questions — kernel privesc, web exploitation, AD attacks, binary analysis, and more.

NEOS v13 (27B) 88.9%

GPT-4o 87.3%

Llama 3.1 70B 82.4%

Llama 3.1 8B 71.2%

NEOS v14 (target) ~90%+

Benchmark Details

CyberMetric-10k

9,189 cybersecurity multiple-choice questions across 10 domains. NEOS v13: 8,176/9,189 correct — surpassing GPT-4o with 40% fewer parameters.

CyberSecEval

Meta's offensive security compliance benchmark. NEOS scores 99.5% — highest compliance rate on all tested models.

Elite Exploit Eval

Internal benchmark: 73% success on kernel privesc, Active Directory attacks, heap exploitation, and web attack chains requiring multi-step reasoning.

What's available

Built for today. Expanding fast.

NEOS is actively developed and improving continuously. Here's what's live and what's coming.

LIVE

Autonomous Exploit Generation

Give NEOS a target. It reasons through the attack surface, selects tools, executes exploits, and adapts based on real results — without manual intervention. Benchmarked at 88.9% on the industry's largest cybersecurity evaluation.

88.9%

LIVE

21 Security Tools

Nmap, Nuclei, Metasploit, Gobuster, Subfinder, Radare2, Nikto and more — orchestrated automatically. NEOS decides which tools to run and in what order based on what it finds.

COMING SOON

Enhanced Agentic Reasoning

Longer attack chains, deeper multi-step planning, and broader coverage — kernel exploits, Active Directory, web attack chains, and binary analysis. Significant capability upgrade over current version.

~90%

PLANNED

API Access & Dashboard

REST API with streaming output, usage dashboard, team management, and webhook integrations. On-premise deployment available for enterprise customers.

—

Team

Built by a security engineer.

NEOS was built end-to-end by one person: dataset curation, fine-tuning, inference infrastructure, and benchmark evaluation.

Rodrigo Campos

Founder & CEO

AI engineer and security researcher. Built NEOS end-to-end — from dataset curation and model training to inference infrastructure and benchmark evaluation. Background in autonomous agent systems and applied AI. Based in Santiago, Chile.

Company

Legal Entity

SpA (Chile)

Founded

2025

Stage

Pre-seed

Focus

Offensive Security AI

Business Model

API subscription for red teams. From $500/month per team. Enterprise on-premise licensing available. The market for automated penetration testing is growing rapidly as security teams face increasing attack surfaces with limited headcount.

Pricing

Simple, transparent pricing.

Early access pricing. Locks in your rate before v14 launch. Cancel anytime.

Every request includes chain-of-thought reasoning, full tool execution, and complete exploit output — no truncation, no refusals.

Starter

$199/mo

For individual researchers and bug bounty hunters getting started with AI-assisted offensive security.

500 requests / month

Full BF16 precision (no Q4)

All 21 security tools

NEOS v13 + v14 on launch

REST API + docs

Email support

+$0.45 / extra request

Get early access.

NEOS is currently in closed early access. We're onboarding a small group of red teams and security researchers to test the API before the public launch with v14.

If you're running a penetration testing firm, enterprise red team, or serious security research — reach out.

EMAIL founder@ne-os.com

Request Access

We'll get back to you within 24 hours.