Polyrouter

What Polyrouter is

Polyrouter is a free, open-source LLM gateway that provides one endpoint for multiple model providers. Requests can be routed by explicit model selection, header rules, routing tiers, or optional automatic routing layers. Ordered fallback chains and per-provider circuit breakers help keep requests moving when a provider degrades. Spend budgets can apply globally or per agent over daily, weekly, or monthly periods, with alerts or request blocking at a threshold. Polyrouter supports OpenAI- and Anthropic-compatible APIs and translates between the two protocols. Providers can be connected with API keys, OAuth subscriptions, compatible custom endpoints, or self-hosted models such as Ollama, LM Studio, and llama.cpp. Any client that accepts a configurable base URL and API key can connect. Request logs contain metadata by default; capturing prompts or responses is an optional self-hosted setting that starts off disabled. Provider and notification credentials are encrypted at rest, and user-supplied URLs are checked against private and metadata address ranges. Self-hosting requires Docker with Compose v2 and runs the SPA, API, and proxy alongside PostgreSQL 16 and Redis. Polyrouter is licensed under AGPL-3.0. Managed cloud hosting is described as coming soon.