ZRouter.

THE SUPER INTELLIGENCE ROUTER

Every request, routed
by how hard it is.

Point your app at one ZRouter endpoint and one virtual model name. ZRouter scores every request in under a millisecond and routes it to the right tier of models you choose. Change providers any time; your app never notices.

Create your account Read the API guide
WHY IT EXISTS

Spend more time building.
Less time managing connections.

Switching AI providers usually means a new SDK or endpoint, new env vars, a new model string, and a redeploy. Adding a fallback provider means all of that twice.

With ZRouter, your app only knows one endpoint and one virtual model. Providers, weights, fallbacks, and schedules live behind that name, and you change them in the dashboard, zctl, or MCP. We translate each request for the provider that serves it.

WHAT MAKES IT DIFFERENT

Your keys and your routing
stay yours.

Four choices that shape how ZRouter works.

Manage it where you work

Use the dashboard for charts and the playground, zctl for terminal commands and scripts, or MCP to let an AI assistant manage your account. Your account controls follow you across all three.

Connect your assistant

Your provider keys, protected

Saved keys are encrypted and bound to your account and provider. Each request runs on isolated clients built from your keys, and works with your virtual models across every provider you saved a key for. It never uses platform keys or the shared response cache. The provider bills you directly, and our routing fee is separate.

How BYOK works

Routing for every customer

Build private virtual models with fallback, weights, and session keeping. Point your application at a stable selector, then adjust its targets from the dashboard, CLI, or MCP.

Virtual models

One API across many providers

OpenAI-compatible chat, Responses, messages, and embeddings reach 34 provider types, from hosted APIs to engines you host yourself such as vLLM, Ollama, and llama.cpp.

See the providers

01 / CLARITY

See what your
application is doing.

Every request leaves a trail you can read. Follow token usage by model, open your request logs, and compare estimated provider cost with what was actually debited from your balance.

The usage overview: a live token throughput chart over 30 days, with cards for input, output and total tokens, total requests, estimated cost, prompt cache rate, and provider status.
Usage overview from a ZRouter workspace.

What you see in request logs depends on the workspace's retention and body-capture settings.

02 / TRUST

Your keys stay yours.

Save a provider key and it is encrypted with AES-256-GCM. The API never returns it.

Requests made with your key run on an isolated provider client, and the provider bills you directly. Any ZRouter routing fee is shown separately.

BYOK setup

03 / SIMPLICITY

Pay for what you use.

Add $5 or more of prepaid credit at a time through Stripe-hosted Checkout. There is no subscription, and every API key in your account draws on one balance.

See token pricing

STAY IN CONTROL

Decide who can do what,
and how much.

Give each application its own key and manage spending and request rules within your account.

Account API keys

Create a named key for each app or teammate. Copy the secret once, and revoke any key the moment you stop trusting it.

Personal budgets

When enabled, set budget rules for your own account and watch spend estimates against them. Reset your own counters when you need to.

Rate limits

When enabled, set rate-limit rules for your account. Changes and resets only ever apply to your rules, never another customer's.

ROUTING THAT STAYS OUT OF YOUR CODE

Name a model once.
Change it whenever.

Give your app one stable name. Decide later what sits behind it, without a redeploy.

Fallback and weights

Point a virtual model at several targets, set weights, and reorder them. If one target fails, fallback retries pick up the request. Session keeping holds a conversation on the same target.

Private selectors

Each virtual model gets a selector such as accounts/<account-id>/my-assistant. Paste it into the model field. It also shows up in the playground and in /v1/models for your keys.

Super Intelligence Router

Let ZRouter read each request and pick the model it needs: quick questions to a fast, cheap model, hard reasoning and code to your strongest one. Scheduled, failover, and lowest-cost routing are there too, and all of them work with your own provider keys.

How it decides

FIND IT. TRY IT.

Start in the playground,
not in your editor.

Browse the models your account can use, then send a real request before you write any code.

Model catalog

See which models are enabled for your account and copy the exact identifier to send. Identifiers use the PROVIDER_INSTANCE/MODEL_ID format.

Playground

Pick a model and send a request with an account API key, paid with credits or your saved provider key.

The real JSON

Switch between Chat Completions, Responses, and Messages, and read the exact request and response. Requests appear in your request logs and usage.

The playground with endpoint tabs, a model picker, message controls, and a side panel showing the request JSON.
The playground. Available models depend on your workspace.

PROVIDERS

One gateway.
34 provider types.

The models you can call depend on what this workspace has enabled. Your dashboard shows the live list.

Frontier and hosted APIs

OpenAI, Anthropic, Google Gemini, xAI, Meta, Cohere, DeepSeek, Z.ai, MiniMax, Xiaomi, Kimi Code, Alibaba Bailian, ElevenLabs, ChatGPT subscription

Inference platforms and routers

OpenRouter, Groq, Cerebras, Fireworks AI, Hugging Face, Chutes, Kilo AI, Cloudflare Workers AI, Hetzner, OpenCode Go

Cloud and self-hosted

Azure, Amazon Bedrock, Bedrock Mantle, Google Vertex AI, Oracle, Ollama, vLLM, SGLang, llama.cpp, llm-d

Bring your own keys works with OpenAI, Anthropic, Gemini, Groq, OpenRouter, DeepSeek, xAI, Cerebras, Hugging Face, Alibaba Bailian, Chutes, Cohere, Fireworks AI, Kilo AI, Kimi Code, Meta, MiniMax, OpenCode Go, Xiaomi, and Z.ai. BYOK uses official provider endpoints. Custom base URLs, Azure credentials, and Vertex service accounts are not supported. BYOK guide

A GATEWAY YOUR APP ALREADY SPEAKS

Same request.
New reach.

If your code can call an OpenAI-compatible API, it can call ZRouter. Use the ZRouter URL below with your account key, pick a model from your dashboard, and send.

Send your first request
chat-completions.sh
curl 'https://zrouter.si/v1/chat/completions' \
  -H "Authorization: Bearer YOUR_ZROUTER_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "PROVIDER/MODEL_FROM_YOUR_DASHBOARD",
    "messages": [
      {"role": "user", "content": "Hello, ZRouter"}
    ],
    "max_tokens": 256
  }'
POST/v1/chat/completionsOpenAI-compatible

WHAT YOU CAN CALL

The endpoints
your account supports.

ZRouter translates supported requests for each provider and returns a familiar response. Model capabilities vary, so check the API guide for the details.

Endpoint support
POST/v1/chat/completionsChat + streaming
POST/v1/responsesSupported text requests
POST/v1/messagesMessages interface
POST/v1/embeddingsText embeddings
GET/v1/modelsAvailable models
GET/v1/usageScoped usage

Prepaid accounts support metered text requests. Media, tools, background jobs, and lifecycle endpoints are restricted.

AN HONEST FIT

Know what fits
before you sign up.

We would rather you know the limits now than find them in production.

A good fit

  • Apps and scripts that use chat, Responses, messages, or embeddings
  • Chat interfaces such as Open WebUI and LibreChat, and n8n workflows
  • People who want one balance, one usage view, and one list of keys

Needs a standalone gateway

Prepaid accounts reject tools, media, background jobs, and stored response continuation. Routing model requests for coding agents such as Claude Code, Codex, Cursor, and Cline needs a standalone gateway. They can separately use the account MCP server to manage your ZRouter account.

Understand standalone gateways

Integrations

Setup guides for coding tools, chat interfaces, workflows, and SDKs, each with its billing requirements spelled out.

Browse integrations

WORK WHERE YOU BUILD

One account.
Three ways to manage it.

Use the dashboard for a visual view, zctl for repeatable commands, or MCP to work with your AI assistant.

See it in the dashboard

Compare usage, inspect available request logs, manage credits, and test models in the playground. Start here when you want to see the whole account.

Open the quickstart ↗

Automate it with zctl

Create and edit virtual routers, manage keys, and review usage from your terminal. JSON output and router exports make account tasks easier to repeat in scripts.

Install the zctl CLI ↗

Ask your agent through MCP

Connect Claude Code, Cursor, or another compatible MCP client. Ask about usage, investigate errors, and manage routers, API keys, and enabled account limits.

Connect an MCP client ↗

CLI and MCP management access is limited to your account. Connecting MCP does not change your agent's inference provider. See each tool's integration guide for model-request support.

YOUR NEXT REQUEST STARTS HERE

Build with AI.
Keep the controls.

One account for model access, routing, and usage. Manage it from your dashboard, terminal, or AI assistant.

Create your account Read the quickstart No card needed to create an account.