vex-router

Intelligent LLM Routing and Observability for the VEX Protocol.

vex-router provides a high-performance routing layer for AI agents, enabling dynamic model selection based on cost, latency, quality, and adversarial role detection.

Features

• Intelligent Routing - Route queries based on quality (GPT-4o/Sonnet) or cost (Mini/Haiku).
• Adversarial Detection - Automatically upgrades routing quality when red-teaming/adversarial roles are detected in the system prompt.
• Semantic Caching - Built-in vector-based caching to reduce redundant LLM calls and latency.
• Observability - Detailed metrics for tracking cost savings, latency percentiles, and request precision.
• OpenAPI Support - Native utoipa support for integration with the VEX API and Swagger UI.

Installation

[dependencies]
vex-router = "0.1"

Quick Start

use vex_router::{Router, RouterBuilder, RoutingStrategy};

#[tokio::main]
async fn main() {
    let router = RouterBuilder::new()
        .strategy(RoutingStrategy::CostOptimized)
        .build();

    let result = router.ask("Calculate the risk profile for this audit.").await;
    println!("Response: {:?}", result);
}

License

Apache-2.0 License - see LICENSE for details.