chinchilla-llm-router

Installer for llm-router — an OpenAI v1-compatible LLM router with logical models, fallback/round-robin routing, and automatic rerouting.

This package is a zero-dependency Node.js bootstrap (Node ≥ 18). It downloads the prebuilt Go binary for your platform from GitHub Releases, verifies it (sha256), installs it under ~/.llm-router/, and generates config.yaml + .env for you. The router itself is a standalone Go binary — Node is only used to install and launch it.

Quick start

npx chinchilla-llm-router

Interactive wizard: server (host/port/auth) → credentials (OpenAI, local OpenAI-compatible, Azure) → logical models → defaults → writes config. Secrets go only to .env (chmod 600); config.yaml uses ${VAR} references.

Then:

npx chinchilla-llm-router run      # start the router in the background (returns immediately)
npx chinchilla-llm-router status   # is it running?
npx chinchilla-llm-router stop     # stop it
npx chinchilla-llm-router doctor   # check binary, config, running server
npx chinchilla-llm-router version  # installer + binary versions

run starts the router detached and returns as soon as /healthz answers; running it again restarts the server (reboot). run --foreground blocks the terminal instead (for debugging). A detached server logs JSON lines to ~/.llm-router/logs/llm-router.log.

Non-interactive (scripts / CI)

npx chinchilla-llm-router init \
  --cred vllm:local:http://192.168.18.200:1235/v1:sk-... \
  --model chat:chat:fallback:vllm:qwen3.8-27b \
  --port 8080 --auth

Commands

| Command | What it does | |---------|--------------| | setup (default) | Install the binary, then run the wizard | | install | Install the binary only (skips if present) | | init | Generate config — wizard or flags | | doctor | Check binary, config validity, running server | | run | Start the router detached (restarts it first if already running); --foreground blocks the terminal | | stop | Stop the detached router | | status | Show whether the detached router is running | | version | Print installer + installed binary versions |

Flags: --dir <dir> (install root), --version <v> (binary version), --force (reinstall), --foreground (with run). Environment: LLM_ROUTER_HOME, LLM_ROUTER_VERSION, LLM_ROUTER_RELEASE_BASE (self-hosted releases).

Supported: Linux and macOS, amd64 and arm64. Windows: use WSL or Docker.

Full documentation: https://github.com/DavidCastilloAlvarado/chinchilla-llm-router#installation