一个两人安全团队用AI挖出了代码里的真实漏洞，一笔赏金就赚了25万美元。

现在他们把这套系统开源了，叫Open-Kritt。它不是那种丢个仓库让AI瞎找的玩具，而是一个能把AI代理变成小规模安全团队的编排平台。

大多数人的做法是扔给AI一句“找这个代码里所有漏洞“，结果往往是一堆误报和没法复现的结论。Open-Kritt的思路更细：先让一个代理把代码里所有可达入口点都画出来，再启动一堆独立代理，每个沿着一条路径去追，专门找特定类型的失败点。

每个代理都有自己的隔离容器，在里面可以翻代码、装安全工具、编译项目、跑测试、做模糊测试、甚至直接构建概念验证。原始发现出来后，系统会检查、合并同类项、按严重程度排好序，研究人员从最可疑的开始看。

兼容Codex、Claude Code、OpenAI、Anthropic和OpenRouter。背后团队用它在九个月里挖出了超过20个高危和严重漏洞，还在Firedancer V1审计竞赛里靠纯AI驱动的研究拿过第一，没经人工审查。

https://github.com/Kritt-ai/open-kritt…

Kritt-ai/open-kritt

Source: https://github.com/Kritt-ai/open-kritt

open·kritt

Orchestrate AI agents to find real vulnerabilities in code.

An open-source, self-hosted security research platform that turns focused AI analysis into de-duplicated, ranked findings with configurable validation and enrichment.

Website · Documentation · Getting started · Contributing · Research paper · Discord community · Twitter

What is open·kritt?

Pointing a model at an entire repository and asking it to find vulnerabilities rarely works well. open·kritt takes a focused approach: break the research into small, well-defined tasks, run them across AI agents in parallel, and combine their output into findings you can validate and prioritize.

It is built for security researchers and security-minded developers who want control over their prompts, workflows, model providers, and infrastructure.

What it does

• Build workflows — chain focused prompts into reusable security research playbooks.
• Run scans — analyze remote or local repositories and their dependencies with Codex or Claude Code.
• Verify findings — use post-scripts to validate issues, build proofs of concept, and produce reports.
• Prioritize results — apply custom severity rankers, a consistent finding schema, and automatic de-duplication.
• Bring your own model access — use a Codex login or connect through OpenAI, Anthropic, or OpenRouter.

Built from real security research. The Kritt team has earned over $1,500,000 in bug-bounty payouts under the researcher name Blockian (Immunefi · HackenProof · blockian.xyz · @Kritt_AI). open·kritt is the open-source distillation of the internal project behind that work.

Getting started

You need Git, Docker with Docker Compose, and Node.js 20 or newer. The repository-local CLI has no install step.

git clone https://github.com/Kritt-ai/open-kritt
cd open-kritt
./kritt setup
./kritt start

Open http://localhost:5173 once the stack is running. You only need one model-access option; ./kritt setup guides you through the available logins and API keys. A GITHUB_TOKEN is optional and only needed for private GitHub repositories.

The default ports bind to 127.0.0.1, and the backend does not include application authentication. Keep the stack private.

Tool-enabled agents run as root inside disposable job containers, with writable repository copies and direct internet access so they can install tools, compile targets, run tests, and build proofs of concept. Run open·kritt on a dedicated Docker host or VM; see the threat model before scanning untrusted code.

For prerequisites, manual Docker setup, and provider-specific instructions, read the installation guide and AI provider setup.

Documentation

Preview the documentation locally with Mint:

npm install -g mint
cd docs-site
npm run dev

Open http://localhost:3001 to view the site.

• Product overview
• Run your first scan
• Workflows and prompt steps
• Security and threat model

Community and contributing

Questions and ideas belong in GitHub Discussions. Use GitHub Issues for bugs and feature requests.

Contributions are welcome. Read CONTRIBUTING.md for the development setup, test commands, Conventional Commits, and DCO sign-off requirements.

Please report security vulnerabilities privately by following SECURITY.md, not through a public issue.

License

open·kritt is licensed under the GNU Affero General Public License v3.0.