Holistic Application Security — Offensive & Defensive Open Source Tooling

A complete open-source ecosystem for application security. From AI-powered pentesting to LLM guardrails and structured assessment methodologies.

View on GitHub Explore Projects
Scroll

What We Do

Full-spectrum application
security, open source.

$ nmap -sV scanning... Offensive Security

A pentest agent that actually thinks.

Most AI “pentesters” are chatbots wrapped around a script library. Agent-Smith flips the model—skills teach the methodology, the LLM invents the attack. Self-chaining across 25+ disciplines: web, cloud, AD, AI red-team, white-box review. Nothing else in open source comes close.

Defensive Security

The LLM firewall built on semantic intent.

Seraph isn’t another blocklist. A two-tier proxy runs a semantic allow-list (NeMo + embeddings) that decides what users are even allowed to ask, then hands edge cases to an LLM-as-judge. A defense designed for prompts you haven’t seen yet. Drop-in for OpenAI, Anthropic, Azure, Ollama—zero code changes.

? Methodology

Skills as pattern teachings, not scripts.

25+ chainable slash-command skills covering the full lifecycle—recon, exploit, defend, remediate. Aligned to OWASP ASVS 5.0, LLM Top 10, MITRE ATT&CK, PASTA, STRIDE. Built by the community, pluggable into any AI agent.

Open Source Arsenal

Built to break. Built to protect.

Three projects. One mission: security that moves at the speed of AI—so your defense does too.

OffensiveAI Pentest Agent

The AI that thinks like an attacker.

Point it at a target and watch a full pentest run itself—findings, PoCs, patches, and reports. It builds a live world model of everything it discovers.

25+
Skills
50+
Tools in parallel
100%
Sandboxed
Explore Agent-Smith
agent-smith · world model LIVE
Target Host Endpoint Vuln
Click a node to inspect it.
seraph · llm gateway GUARDING
Prompt in
Tier 1 · semantic allow-list BLOCK
Tier 2 · LLM-as-judge
Upstream model
DefensiveLLM Guardrail Proxy

A firewall that can’t be talked out of its job.

A positive-security allow-list defines what users are even allowed to ask—so there’s nothing to jailbreak past.

2
Defense tiers
0
Blocklists
5+
LLM providers
Explore Seraph
MethodologySlash-Command Library

Security expertise, one slash command away.

25+ chainable commands that turn any MCP-capable agent into a red-teamer. One git clone. No SaaS.

25
Commands
427
ASVS checks
0
Lock-in
Explore Skills
claude code · /skills 25+ commands
/pentester
/codebase
/ai-redteam
/threat-model
/remediate

The Ecosystem

A continuous security lifecycle.

Every tool feeds the next. Methodology informs testing, testing reveals what to defend, defense generates knowledge, and knowledge refines the methodology.

Skills Learn & share
Agent-Smith Attack & discover
Seraph Defend & protect

← Learn. Break. Defend. Repeat. →

From the Blog

Notes from the frontier.

Deep dives on offensive tooling, LLM defense, and the methodology behind the ecosystem.

2026-07-15 Mobile

Agent-Smith goes mobile: Android & iOS pentesting

Point Smith at an APK, an IPA, or a mobile source tree and it runs the whole MASVS/MASTG methodology itself—static and dynamic—then follows the traffic into the backend API.
2026-07-03 Offensive

Skills, not scripts: how Agent-Smith chains an attack

Most “AI pentesters” are chatbots wrapped around a script library. We took the opposite bet—teach the methodology, let the model invent the attack.
2026-06-26 Defensive

A firewall you can’t jailbreak: designing Seraph

Every blocklist eventually fails. Here’s why we built Seraph on a positive-security allow-list instead—and what it means to be immune to prompt injection by design.

 Read all articles

Community

Build with us.

NullPointer is fully open source and community-driven. Whether you're a pentester, security engineer, or developer—there's a place for you. Contribute code, share knowledge, or just hang out.

Join Discord GitHub Org