# vllm-project/semantic-router

A programmable Mixture-of-Models router for heterogeneous LLM inference

Repository: https://github.com/vllm-project/semantic-router
Canonical: https://ross.abutalabs.com/products/semantic-router
Homepage: https://vllm-sr.ai
Language: Go
License: Apache-2.0
License Family: permissive
Topics: mixture-of-models, semantic-router, vllm, ai-gateway, kubernetes, llmrouter, llm, guardrails, inference, pytorch, transformer
Last push: 2026-08-26T19:57:58+00:00

## Health v2 (maintenance only)
Score: 76/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 75, longevity 26
- inputs: {"age_days": 372, "days_push": 7, "days_rel": 89, "gap_med": 75.5, "n_releases_24m": 3}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 5315, forks 834 (observed 2026-08-28T04:09:15.430292+00:00)

## What it is
vLLM Semantic Router is an open-source programmable routing and control layer for building Mixture-of-Models systems across heterogeneous LLM infrastructure. It runs as an Envoy External Processor, exposing a stable OpenAI/Anthropic-compatible endpoint while selecting or composing model paths based on request signals, policies, and preferences.

## Use cases
- route llm requests to the best model based on intent and complexity
- build a mixture-of-models gateway across cloud and local models
- cut llm inference costs by cascading cheap to expensive models
- add guardrails like jailbreak and pii detection to llm traffic
- route requests across heterogeneous gpu and edge compute
- enforce privacy policies so data stays within boundaries
- balance model quality, latency, and cost with slo-aware routing

## When to choose
- you serve multiple llm models and want a single stable API endpoint with intelligent routing
- you need policy-driven routing with safety, caching, retrieval, and verification attached in the request path
- you want to optimize token spend, latency, and quality across a model fleet without hard-coding logic in apps
- you run kubernetes or envoy-based infrastructure and want routing as an observable control plane

## When to avoid
- you serve a single model with no routing or policy needs
- you want a lightweight client-side router rather than an envoy-based gateway service
- you need a fully turnkey managed solution rather than a self-hosted, config-driven system

## Facets
- artifact type: service
- maturity: active
- function: llm-inference, api-gateway, middleware, routing, security, caching, rag, monitoring
- domain: large-language-models, artificial-intelligence, developer-tools, self-hosted, infrastructure-as-code
- platform: go, python, self-hosted, cloud
- tags: mixture-of-models, semantic-router, envoy-extproc, llm-routing, ai-gateway, guardrails, model-cascades, token-economy, vllm, ai-agents, kubernetes, docker, linux, macos

## Member repositories
- vllm-project/semantic-router (main) score 76

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:15.430292+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:58:57.623954+00:00, confidence not recorded.
  - readme: https://github.com/vllm-project/semantic-router (fetched 2026-08-28T04:09:15.430292+00:00, sha ed1052e72345)
  - homepage: https://vllm-sr.ai (fetched 2026-08-29T08:53:51.815559+00:00, sha e89d50d12efd)
  - site_page: https://vllm-sr.ai/docs/intro (fetched 2026-08-29T08:53:51.825686+00:00, sha adabbd4fdf6b)
  - site_page: https://vllm-sr.ai/docs/v0.3/intro (fetched 2026-08-29T08:53:51.827703+00:00, sha 6ccc35a1ccaf)
  - site_page: https://vllm-sr.ai/docs/v0.2/intro (fetched 2026-08-29T08:53:51.829585+00:00, sha 1a926df69272)
  - site_page: https://vllm-sr.ai/docs/v0.1/intro (fetched 2026-08-29T08:53:51.831264+00:00, sha 9d6c2c1228cb)
  - site_page: https://vllm-sr.ai/docs/installation (fetched 2026-08-29T08:53:51.832938+00:00, sha 0e176c017687)
  - site_page: https://vllm-sr.ai/docs/tutorials/signal/overview (fetched 2026-08-29T08:53:51.835087+00:00, sha fd7e69ecd1ba)
  - site_page: https://vllm-sr.ai/docs/tutorials/algorithm/selection/router-dc (fetched 2026-08-29T08:53:51.837820+00:00, sha ddb865c81d0f)
  - site_page: https://vllm-sr.ai/docs/tutorials/algorithm/selection/multi-factor (fetched 2026-08-29T08:53:51.840251+00:00, sha d75a6cece4d1)
- Data as of 2026-08-30T08:39:29.467469+00:00.