Rate Limit Semaphore (services/rateLimitSemaphore.ts)

Per-model concurrency limiter with FIFO queue for managing request rates and concurrency. Use when throttling API calls or LLM requests in memory.

// Source: https://github.com/b3nw/omniroute-slim | License: MIT | Copyright 2026 diegosouzapw
// Provided via justclaw.it.com; keep this notice with the code.
/**
 * Rate-Limit Semaphore
 *
 * Per-model concurrency limiter with FIFO queue for round-robin combo strategy.
 * When a model is at max concurrency, requests wait in a queue instead of failing.
 * When a model hits rate-limits, it's temporarily paused and queued requests wait.
 *
 * All state is in-memory — resets on server restart (by design, since rate-limit
 * windows are typically short-lived).
 */

interface QueueItem {
  resolve: (release: () => void) => void;
  reject: (error: Error) => void;
  timer: ReturnType<typeof setTimeout>;
}

interface ModelGate {
  running: number;
  max: number;
  queue: QueueItem[];
  rateLimite

... (truncated -- full source via MCP)