Claude Proxy

A high-performance Claude API proxy server that supports multiple upstream AI service providers (OpenAI, Gemini, Custom APIs) and provides Load Balancer, multi-API key management and unified portal access. Functional characteristics ˇIntegrated architecture: Back-end integrates front-end, single-container deployment, completely replaces Nginx 🔐Unified authentication: One key protects all entrances (front-end interface, management API, proxy API) Web Management Panel: Modern visual interface that supports channel management, real-time monitoring and configuration Dual API support: Supports both Claude Messages API (/v1/messages) and Codex Responses API (/v1/responses) Unified portal: Access different AI services through a unified endpoint Multiple upstream support: Supports multiple upstream services such as OpenAI (and compatible APIs), Gemini and Claude Protocol conversion: Messages API supports transfer to other AI services through OpenAI-compatible interfaces 🎯Intelligent scheduling: Multi-channel intelligent scheduler that supports prioritization, health checks and automatic blowing Channel orchestration: Visualize channel management, drag and drop to adjust priorities, and check health status in real time 🔄Trace affinity: The same user session is automatically bound to the same channel, improving a consistent experience Load Balancer: Support polling, random, and failover strategies. Claude/Codex Load Balancer does not affect each other Multiple API keys: Multiple API keys configurable per upstream, automatic rotation (recommended failover policy to maximize Prompt Caching) Enhanced stability: Built-in upstream request timeout and retry mechanisms ensure service remains reliable during network fluctuations Automatic retry and key degradation: automatically switch to the next available key when an error such as insufficient quota/balance is detected; if subsequent requests succeed, move the failed key to the end (degradation); if all keys fail, return as the original upstream error Automatic blowing: Based on a sliding window algorithm to detect channel health, automatic blowing is performed if the failure rate is too high, and automatic recovery will occur after 15 minutes Dual configuration: Support for command-line tools and Web interface management upstream configuration Environment variables: Flexible configuration of server parameters through.env files Health check: Built-in health check endpoints and real-time status monitoring Logging system: Complete request/response logging Support streaming and non-streaming responses ˇSupport tool calls 💬Session management: The Responses API supports session tracking and context preservation for multiple rounds of conversations