HotRiceNoodles/CrossLink

What it solves

CrossLink is a universal LLM API gateway that eliminates provider lock-in by providing a single endpoint for multiple AI models. It solves the problem of having to write separate integration code for every different LLM provider's API format, authentication method, and feature set.

How it works

It acts as a proxy between an application and various LLM providers (such as OpenAI, Anthropic, Azure, DeepSeek, and Ollama). The gateway performs bidirectional protocol translation, converting requests and streaming responses between OpenAI and Anthropic formats in real-time. It includes a routing engine that distributes traffic based on strategies like weighted random or least latency, and an error-classification system that distinguishes between transient and persistent failures to manage failover intelligently.

Who it’s for

Developers and organizations that use multiple LLM providers and want a unified interface, resilient routing, and centralized control over costs, security, and tool access.

Highlights

• Bidirectional Streaming Translation: Converts streaming SSE between OpenAI, Anthropic, and the Responses API using a state machine.
• Error-Classified Failover: Uses a DB-backed rule table to distinguish between quota/billing errors and rate limits to prevent blind retries.
• MCP Gateway: Implements the Model Context Protocol with per-tool RBAC and encrypted credentials.
• Pluggable Guardrails: A registry for blocking, logging, or masking content via custom engines (e.g., keyword filters, credential detection).
• Observability: Provides detailed routing stats, token usage analytics, and x-crosslink-fallback-* response headers.
• Compliance Ready: Supports Chinese national cryptography (SM2/SM3/SM4) and air-gapped deployments.