omniroute

OmniRoute is a self-hosted AI gateway that consolidates access to hundreds of AI providers and models behind a single local endpoint. Licensed under MIT, it requires zero configuration and works immediately after installation without any API keys or signups. ## What It Does OmniRoute aggregates 352+ AI providers offering over 1,200 models, including Claude, GPT, Gemini, GLM, DeepSeek, Kimi, and MiniMax. Over 150 of these provider tiers are free, with a cataloged pool estimated at approximately 1.47 billion free tokens per month. The gateway normalizes all providers behind an OpenAI-compatible `/v1` endpoint, so any compatible tool connects without vendor-specific SDKs. ## Key Features - **Auto-fallback routing**: A quota-aware router automatically falls back across four provider tiers (Subscription → API Key → Cheap → Free) when rate limits or failures occur, keeping your workflow uninterrupted. - **Token compression**: RTK and Caveman stacking compression reduces token usage by 15–95% (averaging ~89%) on eligible workloads, helping stretch free quotas further. - **19 routing strategies**: Flexible request routing with circuit breakers, key cooldown, and model lockout protection. - **MCP and A2A support**: Native Model Context Protocol and Agent-to-Agent protocol integration. - **Local-first**: All requests route through `localhost:20128`. Encryption keys use AES-256-GCM; no provider traffic is tunneled through third-party clouds. - **TLS stealth**: Optional proxy-level TLS masking. - **Live telemetry**: Real-time dashboards track usage, quotas, savings, and p95 latency across all connected providers. - **Guardrails and evaluations**: Production-grade guardrails, evals, and thousands of static test declarations. ## Compatible Tools OmniRoute is designed as a drop-in backend for developer and coding tools, including Claude Code, Codex, Cursor, OpenCode, Cline, GitHub Copilot, and others. A pre-wired keyless provider (OpenCode Free) responds immediately on a fresh install using the `auto` model identifier. ## Deployment Options The gateway is available via npm (`npm i -g omniroute`), Docker Hub, Electron desktop app, and PWA, giving users flexibility in how they run it. Quick-start scripts are provided for Python, Node.js, PHP, and cURL. ## Architecture The Smart Router sits between the caller and provider layer, applying compression and selecting providers from the catalog based on quota availability, cost tier, and routing strategy. Provider keys are pooled for fair-share distribution across teams sharing a single subscription. A terms-risk catalog flags 13 providers where users should review conditions before use. ## Community and Contributors Built by 550+ contributors, OmniRoute has translations in 51 languages. Support channels include Discord, Telegram, WhatsApp groups, and a public roadmap targeting v3.9.0 LTS.