Detect what AI framework a website is built with & find its vulnerabilities
AI Infrastructure Fingerprinting | Built-With Detection | CVE Vulnerability Scanner
Quick Start • API • 60+ Frameworks • 中文
AIFV-Scanner is a lightweight web tool that answers two questions:
- "What AI framework is this website built with?" — Identify Ollama, vLLM, Dify, Gradio, ComfyUI, LangFlow, Jupyter, and 50+ more AI/ML components by HTTP fingerprinting.
- "Is it vulnerable?" — Match detected components against 1190+ CVE rules with version-aware analysis, in both English and Chinese.
Think of it as a "Built With" detector + vulnerability scanner, purpose-built for AI infrastructure.
| Scan & Detect | Infrastructure Base | Add Rules |
|---|---|---|
 |
 |
 |
Probe any URL, IP, or CIDR range and instantly identify the AI/ML framework behind it. Supports 60+ components including LLM serving platforms, AI development tools, MLOps infrastructure, and more.
Automatically match detected components and versions against 1190+ known CVE vulnerability rules. Get severity ratings, descriptions, and remediation advice.
SSE-based live progress — see results as each target completes, not after the entire scan finishes.
Full i18n support for the web UI and CVE descriptions. Switch languages with one click.
Add your own fingerprint rules and vulnerability rules at runtime via REST API — no restart needed. YAML-based DSL for both fingerprints and CVE rules.
docker compose up -dSingle 18MB Go binary with embedded frontend. Host network mode for scanning LAN targets.
git clone https://github.com/NadirAI/AIFV-Scanner.git
cd AIFV-Scanner
docker compose up -d| Scenario | How |
|---|---|
| Detect what AI framework a site uses | Enter any URL and scan — fingerprints reveal the tech stack |
| Security audit of AI infrastructure | Scan your LAN/cloud IPs to find exposed AI services and known CVEs |
| Penetration testing reconnaissance | Identify targets running vulnerable versions of Ollama, vLLM, Jupyter, etc. |
| Asset inventory | Batch scan CIDR ranges to discover all AI/ML services in your network |
| Custom rule development | Add your own fingerprint YAML to detect internal or new AI frameworks |
| Environment Variable | Default | Description |
|---|---|---|
FP_DIR |
data/fingerprints |
Fingerprint rules directory |
VUL_DIR |
data/vuln |
CVE rules — Chinese |
VUL_DIR_EN |
data/vuln_en |
CVE rules — English |
LISTEN_ADDR |
:8899 |
Listen address |
| Method | Endpoint | Description |
|---|---|---|
POST |
/api/v1/scan |
Start scan — {"targets":["..."], "language":"en"} |
GET |
/api/v1/scan/:id |
Get task status and results |
GET |
/api/v1/scan/:id/stream |
SSE real-time progress stream |
GET |
/api/v1/summary |
Component + CVE overview |
GET |
/api/v1/fingerprints |
List all fingerprint rules |
GET |
/api/v1/vulnerabilities?name=xxx&lang=en |
List CVE rules by component |
POST |
/api/v1/fingerprints |
Add fingerprint rule (YAML body) |
DELETE |
/api/v1/fingerprints/:name |
Delete fingerprint rule |
POST |
/api/v1/vulnerabilities?lang=en |
Add CVE rule (YAML body) |
DELETE |
/api/v1/vulnerabilities/:name/:cve |
Delete CVE rule |
| Category | Components |
|---|---|
| LLM Serving | Ollama, vLLM, Llama.cpp, HuggingFace TGI, NVIDIA NIM, Triton, Xinference, LocalAI, LiteLLM, FastChat, Text Generation WebUI |
| AI App Platforms | Dify, LangFlow, Flowise, RAGFlow, n8n, AnythingLLM, PraisonAI |
| Chat UIs | Open WebUI, LobeChat, LibreChat, NextChat, ChuanhuGPT |
| ML/AI Dev Tools | Gradio, ComfyUI, Jupyter Notebook/Lab/Server, Marimo, TensorBoard, MLflow |
| MLOps & Infra | Ray, Kubeflow, BentoML, LangFuse, Feast, Dask, KubePi |
| Training | LLaMA-Factory |
| Data | ClickHouse |
Based on AI-Infra-Guard by Tencent Zhuque Lab. MIT License.
检测网站使用了什么 AI 框架，并发现其漏洞
AI 基础设施指纹识别 | Built-With 检测 | CVE 漏洞扫描
AIFV-Scanner 是一款轻量级 Web 工具，解决两个问题：
- "这个网站用了什么 AI 框架？" — 通过 HTTP 指纹识别 Ollama、vLLM、Dify、Gradio、ComfyUI、LangFlow、Jupyter 等 60+ 种 AI/ML 组件。
- "它有漏洞吗？" — 将检测到的组件与 1190+ 条 CVE 规则进行版本匹配分析，支持中英文双语。
可以理解为专为 AI 基础设施打造的 "Built With 检测器 + 漏洞扫描器"。
| 扫描检测 | 基础设施库 | 规则管理 |
|---|---|---|
|
|
|
输入任意 URL、IP 或 CIDR 网段，即时识别背后的 AI/ML 框架。支持 60+ 种组件，涵盖 LLM 推理平台、AI 开发工具、MLOps 基础设施等。
自动将识别出的组件及版本与 1190+ 条已知 CVE 漏洞规则匹配，提供危险等级、漏洞描述和修复建议。
基于 SSE 的实时进度推送 — 每完成一个目标即时展示结果，无需等待全部扫描完成。
Web 界面和 CVE 描述均支持中英文，一键切换。
通过 REST API 动态增删指纹规则和漏洞规则，无需重启服务。指纹和 CVE 规则均采用 YAML DSL。
docker compose up -d单个 18MB Go 二进制文件，前端内嵌。host 网络模式，可扫描局域网目标。
git clone https://github.com/NadirAI/AIFV-Scanner.git
cd AIFV-Scanner
docker compose up -d| 场景 | 方法 |
|---|---|
| 检测网站使用的 AI 框架 | 输入任意 URL 扫描，指纹揭示技术栈 |
| AI 基础设施安全审计 | 扫描局域网/云主机 IP，发现暴露的 AI 服务和已知 CVE |
| 渗透测试信息收集 | 识别运行了存在漏洞版本的 Ollama、vLLM、Jupyter 等目标 |
| 资产盘点 | 批量扫描 CIDR 网段，发现网络中所有 AI/ML 服务 |
| 自定义规则开发 | 编写指纹 YAML 来检测内部或新兴 AI 框架 |
| 环境变量 | 默认值 | 说明 |
|---|---|---|
FP_DIR |
data/fingerprints |
指纹规则目录 |
VUL_DIR |
data/vuln |
CVE 规则目录（中文） |
VUL_DIR_EN |
data/vuln_en |
CVE 规则目录（英文） |
LISTEN_ADDR |
:8899 |
监听地址 |
| 方法 | 路径 | 说明 |
|---|---|---|
POST |
/api/v1/scan |
发起扫描 — {"targets":["..."], "language":"zh"} |
GET |
/api/v1/scan/:id |
获取任务状态和结果 |
GET |
/api/v1/scan/:id/stream |
SSE 实时进度流 |
GET |
/api/v1/summary |
组件 + CVE 概览 |
GET |
/api/v1/fingerprints |
列出所有指纹规则 |
GET |
/api/v1/vulnerabilities?name=xxx&lang=zh |
按组件列出 CVE 规则 |
POST |
/api/v1/fingerprints |
添加指纹规则（YAML body） |
DELETE |
/api/v1/fingerprints/:name |
删除指纹规则 |
POST |
/api/v1/vulnerabilities?lang=zh |
添加 CVE 规则（YAML body） |
DELETE |
/api/v1/vulnerabilities/:name/:cve |
删除 CVE 规则 |
| 分类 | 组件 |
|---|---|
| LLM 推理 | Ollama、vLLM、Llama.cpp、HuggingFace TGI、NVIDIA NIM、Triton、Xinference、LocalAI、LiteLLM、FastChat、Text Generation WebUI |
| AI 应用平台 | Dify、LangFlow、Flowise、RAGFlow、n8n、AnythingLLM、PraisonAI |
| 对话界面 | Open WebUI、LobeChat、LibreChat、NextChat、ChuanhuGPT |
| ML/AI 开发 | Gradio、ComfyUI、Jupyter Notebook/Lab/Server、Marimo、TensorBoard、MLflow |
| MLOps 基础设施 | Ray、Kubeflow、BentoML、LangFuse、Feast、Dask、KubePi |
| 训练 | LLaMA-Factory |
| 数据 | ClickHouse |
基于腾讯朱雀实验室 AI-Infra-Guard，遵循 MIT 许可证。