// kiri-router · openai-compatible · zero-key

Free models.
Any tool. Zero keys.

A lightweight gateway for free LLM inference. Run it locally with one command and point any OpenAI-compatible client at your own gateway — no account, no API key, no credits, no shared rate limits.

checking status… … models

One command starts the gateway on your machine and opens a full admin console (http://127.0.0.1:8082/) with live model testing, connection snippets, and health monitoring. Every request exits from your own IP. The installer detects which runtimes you have and lets you choose.

# macOS / Linux / WSL — detects Python/Node, asks which to use curl -fsSL https://router.kiri.ng/install.sh | sh
# Windows (PowerShell) — detects Python/Node, asks which to use irm https://router.kiri.ng/install.ps1 | iex

Prefer to choose the runtime yourself? Run a specific one directly — you can view the Python script or the Node script before running:

# Python only curl -fsSL https://router.kiri.ng/run.py | python3
# Node.js only curl -fsSL https://router.kiri.ng/run | node

Requires Python 3 or Node.js. If port 8082 is busy the gateway automatically picks a free port and tells you which one it used.

Works with every OpenAI-compatible client:

Cursor Cline Claude Code ZCode Continue OpenAI SDK cURL …anything OpenAI

Heads up: the shared cloud endpoint draws from a free-tier pool that is frequently rate-limited (HTTP 429), especially from data centers and VPNs. It may work for quick tests — for anything serious, use local mode.

Base URL
https://router.kiri.ng/v1
API key
any string (e.g. kiri)
Model
loading…
Loading model catalog…

Catalog refreshes automatically. Status is probed live from the gateway edge.