Product Introduction

What is XAI Router?

You can think of XAI Router (also known as Yugong Router) as a next-generation AI routing gateway that surpasses openrouter.ai in both features and performance. It's an application-layer, transparent AI API routing and management platform with a complete multi-tenant API system, designed to seamlessly connect and unify all major AI models worldwide.

More importantly, it provides unified access to top-tier coding models like Claude Code Max, OpenAI Codex Pro, and Gemini CLI. This means enterprises can easily and securely distribute premier AI-assisted programming capabilities to every developer on their team, addressing a critical business need to boost R&D productivity.

If you think of model providers like DeepSeek, OpenAI, and Anthropic as "power plants," then XAI Router is the intelligent power dispatch engine that integrates a "national power grid" (high-reliability transmission), "regional distributors" (multi-tenant system), and a "smart metering system" (granular management) all in one.

🔑 100% BYOK (Bring Your Own Key) Architecture

XAI Router adopts a pure BYOK approach:

✅ Your Keys, Your Control: Manage all AI Provider Keys yourself (OpenAI, Anthropic, DeepSeek, etc.)
✅ Zero Markup Pricing: Direct integration with official APIs, no additional fees or service charges
✅ End-to-End Encryption: You are the key, full data control with no super-admin access
✅ No Vendor Lock-in: Export or migrate your configuration anytime with complete freedom

Unlike platforms like OpenRouter, XAI Router does not offer a "paid pool" model, ensuring transparent costs and complete data sovereignty.

The core routing gateway is built on the Rust asynchronous runtime, delivering high throughput, low tail latency, and a smaller memory footprint.

Service Models

Self-Hosted XAI Router [1]: A single-tenant architecture (with an encrypted storage space) where a system administrator centrally manages all AI API Keys and can create sub-accounts for resource sharing. It's designed for large to medium-sized enterprises, financial institutions, healthcare organizations, and government agencies with strict requirements for data sovereignty, information security, and compliance, offering a complete on-premises deployment solution.
Cloud Service XAI Control [2]: A multi-tenant architecture (with encrypted virtual spaces) where each registered user gets an independent AI routing instance to manage their own AI API Keys and can create sub-accounts for resource allocation. It's geared towards global developers, providing an out-of-the-box, highly available AI API access service that supports rapid integration and zero-ops deployment.

Core Difference: The fundamental distinction lies in "who controls the AI API Keys." Both models support creating multi-level sub-accounts and flexibly managing AI API resources.

Solution	Architecture	Management Model	Typical Scenarios	Use Case Example
XAI Router [1]	Single-Tenant	Centralized management by a system administrator who controls all AI API Keys and creates sub-accounts for resource sharing.	Enterprises and institutions with strict requirements for data sovereignty, security, and compliance.	The IT department of a major bank deploys a local router. The admin allocates API quotas for DeepSeek/local models and creates sub-accounts for various business units.
XAI Control [2]	Multi-Tenant	Instant independent routing instance upon registration. Users manage their own AI API Keys and can create sub-accounts to share resources.	Global developers, startups, and scenarios requiring rapid deployment.	Developer Alice signs up and gets an instance, adds her DeepSeek/Anthropic/OpenAI keys, and creates sub-accounts for 5 overseas clients. Developer Bob also signs up and gets a completely isolated instance, managing his own keys and sub-accounts without interference.

Whether you need the convenience of the public cloud or the security of a self-hosted deployment, we provide the professional solution that best fits your needs.

Core Features

Unified API Endpoint: All model calls are routed through a single domain, with XAI Router intelligently dispatching requests to the optimal upstream service.
Broad Model Support: Perfectly integrates with all major global AI models, including the full OpenAI suite, Anthropic's Claude family, Google Gemini, DeepSeek, Mistral, Llama, and more. It also supports specialized or non-standard APIs like Claude Code Max, OpenAI Codex Pro, and Gemini CLI for truly unified access.
Intelligent Key Rotation: Monitors key health in real-time and automatically performs rotation, hibernation, and recovery within the key pool to effectively bypass single-key rate limiting bottlenecks.
Granular Rate Limiting: Supports multi-dimensional limits (TPM/TPH/TPD/RPM/RPH/RPD), IP whitelisting, model whitelisting, and real-time usage reporting with visual monitoring.
Automatic Failover: Upstream errors (timeouts, 429s, 5xx) trigger a millisecond-level circuit breaker, automatically retrying and switching to a healthy node with zero impact on the client-side application.
Flexible Model Aliasing: Supports system-level and account-level model aliasing (e.g., gpt-3.5-*=gpt-4o-mini, deepseek-coder=deepseek-chat), allowing for smooth model upgrades or downgrades without code changes.
Cross-Cloud Protocol Translation: A built-in, high-performance adapter layer supports:
- Azure OpenAI Service → Seamlessly convert to OpenAI API / Anthropic API.
- AWS Bedrock → Convert to OpenAI API / Anthropic API, enabling calls to models like Claude using standard API formats without rewriting code.
- Google Cloud Vertex AI → Convert to OpenAI API / Anthropic API for a unified entry point.
Ultimate Stability: The intelligent key pool scheduling algorithm effectively shields your application from common upstream errors like 429 and 5xx, maximizing the success rate of your business requests.
Seamless Compatibility & Native Protocol Support: Not only is it fully compatible with OpenAI and Anthropic API specifications, but more importantly, XAI Router supports calling models directly via their native API protocols, ensuring full feature access and optimal performance. Simply change the base_url and api_key to integrate seamlessly into your existing applications.
Exceptional Performance: Core service logic runs entirely in memory, ensuring ultra-low latency and outstanding high-concurrency handling.
Native Rust Engine: The core service is built on Rust (async runtime), delivering low tail latency and high throughput.
Powerful Account System: Provides flexible, multi-level sub-account management. Create isolated sub-accounts for team members or customers with fine-grained control over their rates, quotas, rate limits (RPM/TPM), and available models.
Granular Cost Control: All spending is transparent and traceable, from a global overview down to individual records. Supports pay-as-you-go billing with prices consistent with official models.
Zero-Trust Security: In the XAI Control multi-tenant system, core data (XAI API Keys and AI API Keys) is stored with end-to-end encryption at a cryptocurrency-grade level. There is no super-admin, and no plaintext sensitive information is ever stored in the database. Even if the entire database were compromised, an attacker could not decrypt the data. We implement a "You are the key" decentralized security policy—your data can only be unlocked by your unique key, fundamentally eliminating the risk of leaks.

How It Works

XAI Router Architecture Diagram

The Key Pool aggregates numerous upstream AI provider API Keys created from different high-tier accounts. This mechanism effectively solves the service bottlenecks caused by insufficient `TOKEN LIMITS` or `REQUEST LIMITS` on a single official account when facing massive request volumes. We typically maintain at least 100 independent upstream accounts for each model type to handle large-scale concurrent requests from developers worldwide.

Whether you are an independent developer, a startup team, or an enterprise user, XAI Router provides a stable, efficient, and controllable AI API service to help you focus on business innovation.