OmniRoute v3.8.54 is an MIT-licensed, self-hosted OpenAI-compatible gateway that aggregates 380 providers and 1,491 unique chat model IDs behind a single endpoint at localhost:20128/v1. It catalogs 471 free-tier model rows across 34 deduplicated recurring pool keys, computing a headline ~1.62B tokens/month from 17 pools with published positive monthly budgets plus five per-model Groq caps; regional-gated quotas like ModelScope (+~6M tokens) are excluded from the headline. The Auto-Combo engine scores candidates on 16 live factors (health, quota, cost, latency, task fit, quality) across 20 routing strategies including quota-weighted (strategy 11), cache-optimized (16), fusion panel-with-judge (19), and pipeline chaining (20). New in this release: headless mode (omniroute serve --headless) running only the /v1/* proxy; model discovery via GET /v1/models and POST /v1/models/query with capability/context filtering; per-key routing control with forcedModel and atomic /api/keys/{id}/access mutations; quota-share scheduling that fairly splits shared-account quota across pooled keys; adaptive admission control with atomic RPM rolling leases; and 12 composable compression engines (LLMLingua-2, Ultra, omniglyph, GCF v3.2) with fidelity gates. The CLI integrates 39 tools (28 code CLIs, 11 agents) via omniroute run/configure, supports remote scoped-token operation, and offers MITM/TPROXY capture for non-proxy-aware CLIs. Guardrails include prompt-injection and credential-masking red-team suites, optional OIDC login, and opt-in int8 vector memory with typed decay. Media endpoints cover /v1/ocr (Mistral OCR), /v1/audio/translations, and generation via Grok Imagine, Novita, ComfyUI, Firefly, ElevenLabs. Deployment targets span npm/Bun global, Docker multi-arch, Electron desktop, macOS Tauri tray (OmniRouteTray), Android Termux, PWA, and VS Code Copilot Chat via OmniCopilot extension. Radar opt-in overlay adds a mutable ~3B token ceiling; all catalog figures are re-audited biweekly.