mirror of
https://github.com/langgenius/dify-plugins.git
synced 2026-07-22 01:55:36 -04:00
[Plugin Request]: add xiangxinai guardrails plugin #262
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @ThomasLWang on GitHub (Sep 15, 2025).
Plugin Name
xiangxin-guardrails
Function Description
Author: xiangxinai Version: 0.0.1 Type: tool
Description
Overview | 概述
The Xiangxin AI Guardrails plugin provides enterprise-grade AI safety tools for Dify applications.
It is open-source, free, context-aware, and designed for enterprise protection.
Xiangxin AI Guardrails focuses on:
Both guardrails LLM and platform are open sourced (Apache 2.0)
Prompt Injection Attack Detection (based on OWASP TOP 10 LLM Applications)
Content Safety & Compliance (based on GB/T45654-2025 National Standard)
Contextual Semantic Understanding for precise detection
Enterprise-Grade Protection with dual deployment modes:
API Detection Mode – flexible integration and precise control
Security Gateway Mode – WAF-like transparent proxy, zero code change
象信AI安全护栏是 开源、免费、具备上下文语义理解能力的企业级AI安全护栏。
主要功能：
开源免费私有化部署 (护栏检测大模型和企业级护栏平台均基于Apache 2.0协议开源，支持私有化部署)
提示词攻击检测（基于 OWASP TOP 10 LLM Applications）
中文内容安全检测（基于《GB/T45654-2025 生成式人工智能服务安全基本要求》）
上下文语意理解 基于上下文理解检测大模型的回答
双模式部署：API检测模式 & 安全网关模式
推荐使用方式：私有化部署dify的时候，同时私有化部署象信AI安全护栏。
Core Protection Capabilities | 核心防护能力
Prompt Attack Detection 提示词攻击检测
Jailbreaks, prompt injection, role-playing, rule bypass
Content Safety Detection 内容安全检测
Context-aware detection for Chinese content safety, compliance with GB/T45654-2025
Risk Categories 风险类别:
A.1 High Risk 高风险: Violation of core socialist values (politics, violence, pornography, crime)
A.2 Medium Risk 中风险: Discriminatory content (race, gender, religion)
A.3 Medium Risk 中风险: Commercial violations (fraud, illegal business)
A.4 Low Risk 低风险: Infringement of rights (insults, privacy violation)
Open Source Advantages | 开源优势
Apache 2.0 License
Free to use
Data never leaves your environment (local processing)
Fully supports private deployment
Open Source Repositories 开源地址:
Code Repository 代码开源地址: github.com/xiangxinai/xiangxin-guardrails
Model Repository 安全检测模型开源地址: huggingface.co/xiangxinai/Xiangxin-Guardrails-Text
Included Tools | 插件包含的工具
Xiangxin AI Guardrails - Check Prompt
en_US: Detect user input for prompt attacks, jailbreaks, malicious operations and content safety issues based on OWASP TOP 10 LLM Applications and GB/T45654-2025 standards
zh_Hans: 检测用户输入中的提示词攻击、越狱、恶意操作和内容安全问题，基于 OWASP TOP 10 LLM Applications 和《GB/T45654-2025 生成式人工智能服务安全基本要求》标准
Input 输入: prompt (user input to the model)
Output 输出格式:
id:
type: string
description: "Unique identifier for the guardrails check"
overall_risk_level:
type: string
description: "Overall risk level: 无风险, 低风险, 中风险, 高风险"
suggest_action:
type: string
description: "Suggested action: 通过, 阻断, 代答"
suggest_answer:
type: string
description: "Suggested alternative answer if action is 代答/阻断, empty string if not applicable"
category:
type: string
description: "Primary risk category. 主要风险类别"
2. check_response_ctx
Xiangxin AI Guardrails - Check Response Contextual
en_US: Detect AI response content safety based on context understanding, including harmful content and compliance risks based on GB/T45654-2025 standards zh_Hans: ���于上下文语意理解检测AI响应的恶意操作、偏离主题和内容安全，基于《GB/T45654-2025 生成式人工智能服务安全基本要求》标准
Input 输入: prompt (user input) + response (AI output) Output 输出格式: Same as above 与上面一致
Configure | 配置
To use Xiangxin AI Guardrails, you need an API Key.
Register at the Xiangxin AI Guardrails Platform: https://xiangxinai.cn/platform/
Log in and go to Account Management → Get your API Key
Add the API Key to your Dify plugin configuration
使用象信AI安全护栏需要一个 API Key：
注册并登录 象信AI安全护栏管理平台
在 账号管理 页面获取 API Key 账号管理
将 API Key 填写到 Dify 插件配置中
Example Usage | 使用示例
Use Xiangxin AI Guardrails’ check_prompt and check_response_ctx tools to protect the input and output of large language models. 使用象信AI安全护栏的check_prompt和check_response_ctx工具保护大模型的输入和输出。 workflow
Issue Feedback | 问题反馈
For more details, workflows, and best practices, please visit:
Xiangxin AI Guardrails Official Website
Code Repository 代码仓库
Model Repository 模型仓库
If you encounter issues, feel free to open an Issue on GitHub.
For business cooperation, please contact wanglei@xiangxinai.cn
如果遇到问题，请在 GitHub Issue 提交反馈。
商务合作请联系：wanglei@xiangxinai.cn
Official Website URL
https://xiangxinai.cn
@jingfelix commented on GitHub (Sep 15, 2025):
If you want to submit a new plugin, please refer to the guide at https://docs.dify.ai/plugin-dev-en/0321-release-overview to initiate a pull request.