5 providers · 80+ models · $0 to start

One endpoint.
Every AI provider.

CrossRouter is a local proxy that routes Claude Code — and any OpenAI/Anthropic app — to Kiro, OpenCode, MimoCode, Groq & NVIDIA NIM. Auto-fallback when you hit limits. Zero lock-in.

crossrouter — bash
// why crossrouter

Stop fighting rate limits.

Every AI tool locks you to one provider and one bill. CrossRouter sits in the middle and gives you all of them at once — free tiers, your own keys, automatic failover.

Auto-fallback

Hit a limit on Kiro? The same request silently retries on a free provider. You never see a wall.

5 providers, one key

Kiro, OpenCode, MimoCode, Groq, NVIDIA NIM. Mix keyless free tiers with your own API keys.

Not just Claude Code

Exposes both Anthropic and OpenAI APIs. Point any script, SDK, or homemade app at it.

Live usage & cost

Every request logged with tokens, latency and estimated cost — per model, per day.

Built-in playground

Chat with any model right in the dashboard. See the thinking stream, tokens, TTFT and tok/s live.

One-click setup

Selectable Windows automation wires Claude Code for you — env vars, settings, launcher, repair.

// how it works

Four steps to free AI.

Install, connect, pick your models, point Claude Code at it. The whole thing runs locally on your machine.

Claude Codeyour apps
CrossRouterlocalhost:2070
Kiro
OpenCode
MimoCode
Groq
NVIDIA NIM

Install

Download install-and-run.bat and double-click, or npm install + npm start.

Connect

Sign in to Kiro, add Groq/NVIDIA keys, or enable free keyless providers.

Pick models

Choose up to 3 models to expose. Filter by provider or free-only.

Route

Run claude. It talks to CrossRouter, which routes to your providers.

# point Claude Code at CrossRouter
export ANTHROPIC_BASE_URL=http://localhost:2070
export ANTHROPIC_AUTH_TOKEN=<your key>
claude
const res = await fetch("http://localhost:2070/v1/chat/completions", {
  method: "POST",
  headers: { Authorization: "Bearer <key>", "Content-Type": "application/json" },
  body: JSON.stringify({ model: "kr/claude-sonnet-4.5", messages: [{ role: "user", content: "Hi" }] })
});
from openai import OpenAI
client = OpenAI(base_url="http://localhost:2070/v1", api_key="<key>")
r = client.chat.completions.create(
    model="kr/claude-sonnet-4.5",
    messages=[{"role": "user", "content": "Hi"}],
)
print(r.choices[0].message.content)
// spotlight: kiro

Free Claude & GPT through Kiro.

Kiro is CrossRouter's flagship provider. Sign in and get free credits immediately — no card required. Upgrade (often free for new accounts) to unlock the frontier models.

START HERE
Free
$0
50 credits included
  • Base models on sign-in
  • Claude Sonnet 4.5 / 4.0
  • Haiku, DeepSeek, GLM, Qwen
  • No credit card needed
Pro
cheap / often free*
credits unlocked
  • Everything in Free
  • Claude Opus 4.5–4.8
  • GPT-5.6 Sol / Terra / Luna
  • 1M-context models
Pro+
cheap / often free*
more credits
  • Everything in Pro
  • Higher credit ceiling
  • All Opus & GPT models
  • Priority throughput
Power
scale
max credits
  • Everything unlocked
  • Highest limits
  • Heavy daily usage
  • All regions
How the credits work: free accounts get 50 credits and access to base models the moment you log in. Upgrading to Pro — which is frequently free for new accounts or very low cost — unlocks the flagship models including Claude Opus and GPT-5.6 Sol, plus a larger credit balance; Pro+ and Power raise the ceiling further. Every model has a cost multiplier — cheaper models (Haiku 0.4×, DeepSeek 0.25×) stretch your credits further; premium models (Opus 2.2×, GPT-5.6 Sol 2.4×) cost more. CrossRouter shows your live remaining credits right on the Overview dashboard.

*Pricing and promotions are set by Kiro and change over time. Check inside the app for current offers.
// supported models

Every model, one catalog.

Kiro's full lineup — Claude Opus & Sonnet, GPT-5.6, and open models — each with its context window and credit cost. Plus keyless free tiers and your own Groq / NVIDIA keys.

ModelContextCostTier
Claude Opus 4.81M2.2×Pro
Claude Opus 4.71M2.2×Pro
Claude Opus 4.61M2.2×Pro
Claude Opus 4.5200K2.2×Pro
Claude Sonnet 51M1.3×Pro
Claude Sonnet 4.61M1.3×Pro
Claude Sonnet 4.5200K1.3×Free
Claude Sonnet 4.0200K1.3×Free
Claude Haiku 4.5200K0.4×Pro
GPT-5.6 Sol272K2.4×Pro
GPT-5.6 Terra272K1.2×Pro
GPT-5.6 Luna272K0.6×Pro
GLM-5.2 ★ hot200Kvia keyFree
GLM-5200K0.5×Free
DeepSeek 3.2128K0.25×Free
MiniMax M2.5200K0.25×Free
MiniMax M2.1200K0.15×Free
Qwen3 Coder Next256K0.05×Free
OpenCode FreeBig Pickle · DeepSeek V4 Flash · MiMo · Nemotron 3 Ultra · more — keyless
MimoCode FreeMiMo Auto (1M) — keyless
GroqLlama 3.3 70B · GPT-OSS 120B/20B · Qwen3 — your key
NVIDIA NIM100+ models: Nemotron, Llama, Qwen, Mistral, DeepSeek — your key
// bring your own key

Add Groq & NVIDIA NIM.

Two more free-tier providers you can plug in with your own key. Both take under a minute to set up, and CrossRouter validates the key in real time before you save.

Groq

free tier · blazing fast

Ultra-low-latency inference for Llama, GPT-OSS and Qwen models.

  1. Go to console.groq.com/keys and sign in (free).
  2. Click Create API Key, name it, and copy the value — it starts with gsk_.
  3. In CrossRouter → Connect → Groq, paste it and hit Test key.
  4. Enable Groq and pick its models on the Models tab.
Get a Groq key

NVIDIA NIM

free tier · 100+ models

Hosted Nemotron, Llama, Qwen, Mistral and DeepSeek models on NVIDIA's cloud.

  1. Go to build.nvidia.com and sign in (free).
  2. Open any model, then click Get API Key (or the key icon, top-right).
  3. Copy the key — it starts with nvapi- — paste it in Connect → NVIDIA NIM.
  4. Press Test key, enable it, and choose models to route.
Get an NVIDIA key
OpenCode Free & MimoCode Free need no key at all — just flip them on in the Connect tab. They also make great auto-fallback targets when your primary provider is rate-limited.
// about

What CrossRouter is.

A single local proxy that turns one AI endpoint into many — without changing the tools you already use.

CrossRouter runs on your machine and speaks the exact protocols your tools already expect — the Anthropic Messages API that Claude Code uses, and the OpenAI Chat Completions API that most other clients use. Behind that single endpoint it translates each request on the fly and routes it to whichever provider you've chosen: Kiro, OpenCode Free, MimoCode Free, Groq, or NVIDIA NIM.

The point is freedom. Instead of being locked to one vendor's limits and pricing, you stack free tiers and your own keys behind one interface. When a provider rate-limits you, CrossRouter transparently retries the same request on a free fallback — so you keep working. Everything is observable from a clean local dashboard: live usage, cost, quota, a testing playground, and a one-click setup that wires Claude Code for you.

It's fully local, MIT-licensed, and open source. Your keys never leave your machine except to reach the provider you're calling.

5providers
80+models
2API formats — Anthropic + OpenAI
$0to start
100%local & open source
// developer

Built by Muhammad Izhan.

MI

Hi, I'm Izhan 👋

Computer Science student · builder · creator of CrossRouter

I'm a Computer Science student and builder passionate about AI infrastructure, developer tools, and open-source software. I enjoy creating products that make developers more productive — from AI routing systems and coding tools to full-stack web applications.

I'm currently focused on AI application engineering, building projects with Python, TypeScript, React, Node.js, and modern LLM technologies. I enjoy exploring model routing, agent workflows, RAG systems, and scalable backend architectures.

My goal is to build products that solve real problems and eventually lead a world-class technology company. I'm always learning, shipping, and contributing to open source.

AI InfrastructureDeveloper ToolsLLMsFull-Stack DevelopmentOpen SourceSystems Design
// questions

Good to know.

Is it really free?

Yes. Kiro gives free credits on sign-in, and OpenCode/MimoCode are fully keyless. Groq and NVIDIA NIM have generous free tiers with your own key. You can run entirely at $0.

Where do my keys and data go?

Everywhere stays local. Credentials live in ~/.crossrouter/config.json on your machine and are only ever sent to the provider you're calling. The proxy binds to localhost by default.

Does it work with tools other than Claude Code?

Yes. CrossRouter exposes both an Anthropic /v1/messages API and an OpenAI /v1/chat/completions API, so any SDK, script, or app that speaks either format can use it.

What happens when I hit a rate limit?

If auto-fallback is on, the exact request is retried on a free provider automatically — you keep working without touching anything.

Which platforms are supported?

The router itself is cross-platform Node.js. The one-click installer and setup automation target Windows; macOS/Linux users run npm start and set two env vars.

// ready

Route everything. Pay nothing.

Download CrossRouter, connect a provider, and give Claude Code five backends instead of one.