Free-form agents, hard guardrails
The models decide what to probe. Turn budgets, loop detection, and scope on every call keep them in line.
Your self-improving
cybersecurity team.
Research Preview
Maintained by the Swiss Applied AI & Cybersecurity Research Lab.
Paste into your coding agent
Cursor
Claude Code
Codex
Gemini CLI
GitHub Copilot
OpenCode
Install the open-source 0sec security CLI and use it in this project.
1. Install it: curl -fsSL https://raw.githubusercontent.com/0sec-labs/0sec/main/install.sh | bash
2. Verify it: 0 --help
3. Read https://0.security/llms.txt to learn its commands, then run it directly. For example: 0 review . or 0 scan --target <url> --scope ./scope.json. Add --format sarif for machine-readable output.
The Swiss Army knife for offensive security.
Built to be extensible. Create and add your own tools as you go.
- NetworkIdentity, AD, cloud and federation
- Web applicationsSQLi, IDOR, XSS, SSRF, auth bypass
- APIsTenant isolation, BOLA, business-logic abuse
- AI & LLMsPrompt injection, jailbreaks, MCP tool abuse
- Source codeInjection, auth, deserialization, memory safety
- DependenciesSupply chain, malicious packages, CVE replay
- RuntimeContainer escape, sandbox and VM boundaries
- Operating systemPrivilege escalation, service and library flaws
- Kernel0-day hunt into the layer beneath it all
Pentest a web app, AI/LLM endpoint, or MCP serverblack-boxwhite-box
scanevalagent-assureReview source, packages, C/C++, or a kernel treewhite-box
reviewfile-reviewdeep-reviewauditRecon an attack surfaceblack-box
reconjs-reconnpm-discoveryintelHunt a bug class or kernel variantswhite-box
huntkernelcveWork with the evidence a run produced
findingshistoryresumereplayverifytimelinediscloseGenerate and re-test a source fixwhite-box
fixAssess identity and AD posture, read-only and offline
identityadgraphentragraphAnalyze a compiled binary, no sourceblack-box
0verseIntegrate it
mcp-serverconsoletuidashboardIt has to work twice.
A blind verification agent re-exploits every finding. What it can't reproduce never ships.
Reproduce before trust
A blind agent re-exploits each finding from the PoC alone. What it can't reproduce is dropped.
Triage before verify
Class oracles and a second scanner cut the noise before the expensive step runs.
Bring your own model
Anthropic, OpenAI, Azure, OpenRouter, or local. You hold the key.
