Awesome Infra for AI › LLM Gateways & Proxies

weave-os/router

⭐ 5568 Go added to this list on 2026-09-07 repository created 2026-04-27

Weave Router is a drop-in proxy that sits between AI coding tools or applications and multiple LLM providers, selecting the best model for each individual request rather than applying one model to a whole session. It exposes an endpoint compatible with Anthropic Messages, OpenAI Chat Completions and Gemini native APIs, including streaming, tool calls and vision, and can also reach open-source models (DeepSeek, Kimi, GLM, Qwen, Llama, Mistral) through OpenRouter or any OpenAI-compatible endpoint. Routing decisions are made per "action" (a single upstream API request) using a cluster scorer derived from published research on model routing, implemented as a small on-box ONNX embedder rather than a prompt-based heuristic. The project can be used as a hosted service through a one-command npx installer that wires configuration into Claude Code, Codex, opencode or the pi CLI, or self-hosted as a Docker Compose stack consisting of the router, the scorer, a Postgres database for installations, keys and usage, and an optional dashboard UI. Provider API keys are bring-your-own-key by default and stay on the user's machine, encrypted at rest; in self-hosted mode, prompts go directly from the router to the configured provider rather than through Weave's own infrastructure. The router emits OpenTelemetry traces out of the box, viewable in a bundled dashboard or forwarded to external observability backends such as Honeycomb, Datadog or Grafana. It also supports an optional frozen HMM routing policy as a companion sidecar process. The project targets developers and teams using AI coding agents or LLM-backed applications who want automatic, cost-aware model selection and multi-provider access without hardcoding a single model or provider into their tooling.

https://github.com/weave-os/router

llm-routerai-gatewaymulti-providermodel-routinggo