AI Security Gateway

Appearance

🎉 First Public Beta Release ​

The AI Security Gateway is a unified security platform providing real-time monitoring, policy enforcement, and threat detection for Large Language Model (LLM) APIs, Model Context Protocol (MCP) servers and Agent-to-Agent (A2A) registry. This beta release represents a comprehensive security proxy and monitoring platform for AI infrastructure.

✨ Core Features ​

💰 Budget Limits & Cost Control ​

  • Monthly Budget Limits: Set spending limits per user group/team in USD
  • Configurable Warning Thresholds: Alert when spending reaches configured percentage (e.g., 80%)
  • Flexible Enforcement: Block requests at threshold or continue with warnings
  • Automatic Monthly Reset: Budgets automatically reset at month start
  • Manual Reset: Reset budgets on-demand via API or web interface
  • Real-Time Tracking: Monitor current spending, utilization, and remaining budget
  • Dashboard Metrics: View budget status across all teams
  • Request Blocking: Automatic blocking when budget exceeded (HTTP 402 Payment Required)

🔄 Multi-Proxy Management ​

  • Unlimited Proxy Instances: Run multiple MCP and LLM proxies simultaneously
  • Transport Protocol Support: HTTP, WebSocket, and Server-Sent Events (SSE)
  • Centralized Management: Create, configure, start, stop, and restart via web interface
  • Database Persistence: Configurations, alerts, and audit logs stored in SQLite
  • Per-Proxy Monitoring: Individual health checks and performance metrics
  • Auto-Start Support: Configure proxies to start automatically on system boot

🛡️ Custom Security Policies ​

  • 159+ Detection Rules: Built-in rules across critical, high, medium, and low severity
  • JSON-Based Configuration: Easy-to-customize policy definitions
  • Real-Time Threat Detection: SQL injection, command injection, path traversal, XSS, and more
  • Policy Templates: critical-security, standard-security, mcp-advanced-security, llm-security and many more
  • Per-Proxy Assignment: Different security levels for different use cases
  • Custom Policy Creation: Organization-specific security requirements
  • Advanced Pattern Matching: Regex-based rules with polymorphic detection

💬 System Prompt Injection (Experimental) ​

  • Automatic Injection: Security controls injected into LLM requests automatically
  • Per-Proxy Assignment: Different prompts for different LLM proxies
  • Per-User Group Assignment: Fine-grained control via API keys/groups
  • Security & Compliance: Enforce data protection, compliance, and behavioral guidelines with system prompts injected into requests
  • Audit Logging: Complete logging of all prompt operations

🔐 OAuth Proxy with Dynamic Client Registration ​

  • Dynamic Client Registration (DCR): RFC 7591 compliant automatic client registration
  • Dual Operating Modes:
    • Upstream Mode: Pass-through proxy forwarding tokens to OAuth providers
    • Gateway Mode: Full OAuth 2.1 Authorization Server managing tokens
  • Automatic Discovery: RFC 8414 compliant well-known endpoints
  • PKCE Enforcement: Complete RFC 7636 support with dual PKCE flows
  • Supported Providers: Google, GitHub, Azure AD, Okta, Auth0, GitLab, Keycloak, custom OAuth 2.0/2.1
  • Hybrid Authentication: OAuth tokens AND API keys work simultaneously
  • User Attribution: Full request attribution with identity, email, timestamps, risk scores
  • Secure Token Management: AES-256-GCM encrypted storage with automatic refresh
  • Consent Screens: Optional user consent for OAuth flows
  • Audit Logging: Complete OAuth transaction logging for compliance (SOC2, ISO 27001, HIPAA, GDPR)
  • Zero Configuration Clients: Works with Cursor IDE, Claude Desktop, and other MCP clients

📊 Real-Time Monitoring & Analytics ​

  • Live Dashboard: Real-time metrics, request rates, error rates, security alerts
  • Traffic Analysis: Monitor all requests and responses through proxies
  • MCP Tool Discovery & Control: Automatic discovery with fail-closed security (disabled by default)
  • LLM Token Usage Tracking: Monitor consumption, costs, and usage patterns per proxy
  • WebSocket Integration: Real-time updates without polling
  • Alert Management: Advanced filtering, pagination, status workflow
  • Observability & Monitoring: Langfuse, Prometheus, Grafana, OpenTelemetry, Jaeger integration
  • Integrations & Notifications: SIEM, SOAR, Slack notifications

📈 Observability & Metrics Monitoring ​

  • Langfuse: Send LLM traces to Langfuse for observability and analytics, easily configured from the Settings
  • Prometheus Metrics: Native metrics endpoint (/api/v1/metrics/prometheus)
  • Grafana Integration: Pre-built dashboard queries and panels
  • OpenTelemetry Tracing: Distributed tracing via OTLP protocol
  • Jaeger Integration: Full support for trace visualization and debugging
  • Comprehensive Metrics: Request percentiles (p50, p95, p99), database performance, connection pools, policy stats, WebSocket metrics, system resources
  • Production Ready: Configurable sampling rates, trace propagation, OTLP-compatible backends
  • Zero Configuration: Metrics enabled by default; tracing configurable via environment

🛡️ MCP Vulnerability Scanner ​

  • Runtime Security Analysis: Real-time vulnerability scanning of active MCP servers
  • Tool Permission Auditing: Monitor and validate MCP tool capabilities and access levels
  • API Surface Assessment: Comprehensive scanning of exposed methods and resources
  • Malicious Payload Detection: Identify security risks in tool responses
  • Compliance Validation: Ensure MCP servers meet security standards
  • Policy Integration: Automatic blocking of vulnerable endpoints

🚨 MCP Tool Change & Rug Pull Detection ​

  • Tool Behavior Monitoring: Continuous analysis of execution patterns and capabilities
  • Change Detection: Alert on unauthorized modifications to tool definitions, permissions, behavior
  • Rug Pull Prevention: Detect sudden removal, disabling, or functionality changes
  • Supply Chain Security: Monitor for malicious updates or compromised dependencies
  • Baseline Establishment: Create security baselines and alert on deviations
  • Attribution Tracking: Audit trails of configuration changes
  • Automated Response: Automatically disable or quarantine suspicious tools

🤖 Agent-to-Agent (A2A) Registry ​

  • Agent Registration: Auto-discover via AgentCard URLs or manual JSON registration
  • AgentCard Management: Automatic fetching and updating of capabilities/metadata
  • Access Control: Fine-grained user group-based access with per-agent rate limiting
  • Agent Invocation: Send messages with streaming response support (SSE)
  • Task Management: Track stateful operations (submitted, working, completed, failed)
  • Security Integration: All invocations validated by policy engine with risk scoring
  • Audit Logging: Complete audit trails with user attribution and request/response payloads
  • Real-Time Monitoring: View invocation logs, task states, agent usage analytics

🔄 Cross App Access (XAA) - 🧪 Experimental ​

Okta Identity-JAG token support for cross-application access control:

  • ID-JAG Token Exchange: Validate and exchange Okta Identity-JAG tokens for cross-app authorization
  • Client ID Mapping: Configure mappings between IdP client IDs and resource authorization server client IDs
  • Token Revocation: Revoke ID-JAG tokens individually by JTI or in bulk by subject/IdP provider
  • JWKS Caching: Automatic fetching and caching of JWKS from Okta IdP for efficient token validation
  • Per-Proxy Configuration: Enable XAA on specific proxy instances through the web interface
  • Statistics Dashboard: Real-time monitoring of ID-JAG token usage, JWKS cache status, and client mapping counts
  • Audit Integration: Complete audit logging of XAA events (token exchange, validation, revocation) for compliance
  • Web UI Management: UI components for managing client mappings, viewing stats, and revoking tokens

⚠️ Experimental Feature: XAA currently supports Okta's proprietary ID-JAG token format. This feature is under active development and may change significantly as cross-app access standards evolve. Use with caution in production environments.

📦 Deployment & Distribution ​

Release Packages ​

  • Pre-built Binaries: Linux (amd64/arm64), macOS (amd64/arm64), Windows (amd64)
  • Helper Scripts: install.sh, verify.sh, start.sh (Unix) / start.ps1 (Windows)
  • Service Files: Systemd (Linux), Launchd (macOS) for production deployment
  • Pre-built Frontend: Production-optimized Vue.js app
  • Docker Configuration: docker-compose.frontend.yml and Dockerfile.frontend
  • Security Policies: 7 JSON policy files with 159+ detection rules
  • Configuration Templates: env.example, config.example.yaml
  • Documentation: Complete setup and usage guides

Deployment Options ​

  • Docker Frontend: Nginx-based container serving Vue.js app (recommended)
  • Manual Web Server: Deploy to nginx/apache with reverse proxy configuration
  • Native Binary: Go binary for optimal performance (no Docker required for backend)
  • Hybrid Deployment: Mixed environment support (binary + Docker, manual web server)

🔒 Security Features ​

  • JWT Authentication: Secure token-based authentication with role-based access control (RBAC)
  • OAuth 2.1 Support: Full OAuth integration with Dynamic Client Registration (DCR)
  • Session Encryption: AES-256-GCM encryption for OAuth session tokens
  • Policy-Based Blocking: Real-time threat blocking based on security policies
  • Audit Logging: Comprehensive logging of all security events and user actions
  • CORS Protection: Configurable CORS policies with security headers
  • Rate Limiting: Per-user and per-IP rate limiting to prevent abuse
  • Input Validation: Comprehensive validation of all API inputs
  • SQL Injection Prevention: Parameterized queries throughout the application

📚 Documentation ​

Complete documentation suite included:

  • Installation and setup guides
  • Multi-proxy management documentation
  • A2A Agent Registry guide
  • Security policies and custom policy creation
  • OAuth proxy setup and configuration
  • API reference with 90+ endpoints
  • Alert system and traffic analysis guides
  • Observability and monitoring integration
  • Docker and hybrid deployment guides
  • Troubleshooting and best practices

🧪 Testing & Quality ​

  • Unit Tests: Comprehensive test coverage for core components
  • Integration Tests: End-to-end testing with mock servers
  • Benchmarks: Performance testing suite
  • CI/CD: GitHub Actions workflow with automated testing
  • Code Quality: golangci-lint with security rules enabled
  • Security Scanning: gosec security analysis

⚠️ Known Limitations ​

  • Beta Software: Thorough testing recommended before production deployment
  • Static Analysis Limits: Cannot detect all vulnerability types (e.g., semantic issues, prompt injection)
  • Performance Variability: May vary with large codebases or high-traffic environments
  • MCP Protocol Coverage: Some advanced features may not be fully covered
  • Test Environments: Recommended for testing and development environments

📈 Project Metrics ​

  • 159+ Detection Rules: Across all security policies
  • 90+ API Endpoints: Complete REST API coverage
  • 35+ Audit Event Types: Comprehensive compliance logging
  • 6 Database Models: Efficient data persistence layer
  • 7 Pre-built Policies: Ready-to-use security configurations
  • 3 Transport Protocols: HTTP, WebSocket, SSE support
  • 5 Platform Builds: Linux, macOS, Windows (multiple architectures)

🎯 Next Steps ​

  • Community feedback incorporation
  • Performance optimization based on real-world usage
  • Enhanced security policy templates
  • Additional enterprise integrations
  • Extended observability features
  • Stable release planning

Note: This is a beta release intended for testing and development. While the software has been thoroughly tested, users should conduct their own validation in production-like environments. Community feedback is welcomed and encouraged to help improve the software.

For detailed release information and downloads, see the GitHub Releases page.