Major Mizan Updates: One Platform to Route, Track and Reduce AI Spend

Over the past few months, Mizan has evolved from a simple way of routing model calls into a broader platform for understanding and optimizing AI spend.

One API layer across your AI stack

The Router has grown into a comprehensive integration layer. It now supports 150+ models across 10+ providers. Generate a Mizan API key, top up your wallet, and you have access to OpenAI, Anthropic, Gemini, Bedrock, Groq, Mistral, and more.

Projects and scoped API keys let you organize your usage. Issue one key to your frontend team, another to your backend, another to your ML team — each with separate budgets and spend tracking.

Real-time telemetry captures every call: tokens, latency, cost, model, provider, and project. No batch processing delays. No missing data.

// Simple drop-in integration
const client = new Anthropic({
  apiKey: process.env.MIZAN_API_KEY,
  baseURL: "https://api.app-mizan.com/v1",
});

The Router is OpenAI-compatible, which means teams already using OpenAI SDKs can migrate in minutes.

AI spend, finally in one place

The Analyst is where the full picture emerges. For most organizations, AI spend lives in fragmented places: direct provider APIs, subscriptions through SaaS platforms, enterprise licensing agreements, and shadow spend that finance doesn't see.

Mizan brings all of it together. Connect your OpenAI admin key, and we'll sync their usage data. Do the same for Anthropic. Connect your Gmail or Outlook inbox, and we'll extract and categorize AI subscription receipts automatically.

The result is a complete spend view:

• Total AI spend across all sources
• API spend routed through Mizan
• Subscription spend from provider invoices
• Daily spend patterns and trends
• Provider breakdown by cost and tokens
• Project attribution for cost allocation
• Token and latency tracing per call

This is a key Mizan differentiator. Most gateways only see traffic they route. Mizan's broader FinOps direction means we help you understand the entire AI cost picture, even for spend that never touched our API.

Bringing spend that never touched the Router into Mizan

One of the biggest aha moments for Mizan customers is realizing how much AI spend they're blind to. A team uses OpenAI directly. Another uses an AI SaaS tool with a monthly subscription. A third signs up for Claude through a web interface.

Connecting admin keys from OpenAI, Anthropic, and other providers lets Mizan pull spend data in real time. No credential exposure. No API keys shared. We securely sync usage and charge data to give you complete visibility.

Email receipt sync is even simpler. Authorize Mizan to access your Gmail or Outlook inbox, and we'll automatically extract and categorize receipts from common AI providers. Upload receipts manually if you prefer.

The point: spend visibility doesn't require everyone to route through Mizan. Financial visibility is available to all users, regardless of where AI usage actually happens.

From visibility to optimization

Knowing what you spend is the foundation. Knowing where to cut is the next layer.

Khabir, Mizan's optimization layer, analyzes your real usage patterns and surfaces opportunities:

• Duplicate subscriptions that can be consolidated
• Overspending on more expensive models when cheaper alternatives could work
• Opportunities for response caching to eliminate redundant API calls
• Prompt caching to reduce repeated token consumption
• Usage patterns that suggest agentic caching could help

The goal is never to make arbitrary promises about savings. Instead, Khabir helps you optimize spend while maintaining the quality and latency your applications require.

Better attribution with projects

One of the most common questions from finance teams is: "Who is using this, and why?"

Projects in Mizan solve this. Don't just know that you spent $10,000 on Claude. Know that $4,000 came from your customer support chatbot, $3,500 from your content generation pipeline, and $2,500 from your internal research tool.

Every API call, every synced provider bill, every tracked subscription — all attributed to the project that generated it. Finance can build budgets per project. Engineering can own their team's spend. Leadership can see which initiatives are driving cost.

Drop-in compatibility

Mizan is designed to be adopted incrementally. You don't need to migrate everything at once.

Supported SDKs include:

• OpenAI Python and JavaScript
• Anthropic Python and JavaScript
• Vercel AI SDK

More SDKs are coming. The core principle is that integration should take minutes, not weeks.

Built around a simple idea

Control every dollar.

AI spend is becoming a material cost for organizations. It deserves the same rigor and visibility that companies apply to cloud spend, data spend, and software spend.

Mizan is trying to become the control layer that connects:

• Models and providers
• Applications and usage
• Usage and cost
• Cost and optimization

When those layers are connected, AI spend stops being invisible and starts being managed.

What's next

The foundation is in place. The Router works. The Analyst gives visibility. Khabir offers optimization advice.

We're continuing to build toward deeper optimization, richer enterprise controls, broader integrations and better ways for teams to understand the economics of AI.

If you're thinking about AI spend differently — if you need visibility, control, or optimization — it's a good time to see what Mizan can do.