llm-d-router vs router
LLM inference routing and agent model routing compared
llm-d-router directs LLM inference traffic within model-serving infrastructure, while router provides a proxy that routes agent and application requests across model providers. Their main difference is focus: llm-d-router uses inference signals such as backend load and KV-cache locality, while router scores actions to select a model.
llm-d-router: Route LLM Inference with Envoy and Gateway API
llm-d-router is a Go-based router for LLM inference traffic, using an Endpoint Picker with Envoy and the Kubernetes Gateway API. It routes requests using signals such as model-server load and KV-cache locality, with options for standalone or gateway deployments.
router: Route Agent Prompts with Avengers-Pro Scoring
Weave Router is a Go proxy for Claude Code, Codex, Cursor, and other clients, routing each request across model providers. Its in-process cluster scorer, derived from the Avengers-Pro research, uses an on-box embedder to choose a model per action.
| llm-d-router | router | |
|---|---|---|
| Language | Go | Go |
| License | Apache-2.0 | Apache-2.0 |
| Stars | 422 | 5.6k |
| Forks | 472 | 159 |
| Last analyzed | Oct 20, 2026 | Oct 20, 2026 |
Key differences
- llm-d-router targets inference infrastructure and Kubernetes deployments; router targets agent clients and applications using Anthropic, OpenAI, or Gemini APIs.
- llm-d-router selects endpoints using InferencePool state, including load and KV-cache locality; router uses an in-process cluster scorer derived from Avengers-Pro to select a model per action.
- llm-d-router integrates with Envoy through ext-proc and supports Gateway API deployments; router presents a single endpoint to clients and connects to configured hosted providers or OpenAI-compatible endpoints.
- llm-d-router includes request prioritization, flow control, and a sidecar for coordinating inference stages; router includes route previews, usage and cost analytics, and OpenTelemetry traces.
- Both projects are written in Go and licensed under Apache-2.0. llm-d-router has 422 stars and 472 forks, while router has 5.6k stars and 159 forks.
- llm-d-router is described as active, with regular commits and releases; router has high recent activity but a sparse measured release history, and its Cursor support is labeled early beta.
Choose llm-d-router if you…
- need routing decisions based on inference-backend load or KV-cache locality.
- want to connect model-serving infrastructure through Envoy or Kubernetes Gateway API.
- need request prioritization or coordination across prefill, decode, and encode stages.
Choose router if you…
- want agent clients such as Claude Code or Codex to route requests among configured models through one endpoint.
- need a per-action model scorer and support for Anthropic, OpenAI, or Gemini API clients.
- want route previews, usage and cost analytics, or OpenTelemetry traces.
Compare the numbers side by side, or add more projects
This comparison is generated with AI from the OSRepos analyses of both projects. Always check each project's repository and documentation before choosing.