Skip to content
FoundationAI teamsMaestroZB CloudApply for access

Maestro

The AI model that knows every AI model.

Maestro is the AI model at the heart of ZB Stack. Give it a task, and it makes two choices: which model should do the work, and which cloud should run it. Then it checks what comes back, tries again or goes to a stronger model when the work is not good enough, and learns from every result. This page shows how, one step at a time.

Apply for access Follow one task

How it works

Give Maestro a task.

It breaks the task into steps. It routes each step to the model that does it best. It checks the work, tries again when a test fails, puts it all together, and remembers what worked. So every job is done by the best model for that step, for the price of the cheapest one that can do it well.

What it is

Maestro does not do the work. It knows who should.

Think of an orchestra. Violins, drums, a piano, a flute. Each one plays its part better than anything else on the stage. The maestro plays none of them. The maestro knows which one to call on, and when, and hears when a note is wrong, so the whole piece sounds right.

AI works the same way now. There are thousands of models. One writes the best plan. One writes the best code. One draws the best pictures. One answers a customer better than all the rest. Maestro knows each of them: what it is good at, what it costs, and how fast it works.

We built Maestro by watching millions of real tasks go to real models. We saw which model did each kind of task best, where it ran fastest, and what it cost. We saw which plans worked, and how many tries each one took. Every route, every check and every result went into one model.

  • The models

    300+ AI model endpoints across 14 leading model providers.

  • The clouds

    Big public clouds, in-country clouds, neo centers and shared clouds.

  • The results

    Which plan worked, which model won, how fast, at what cost, and in how many tries.

Two choices

Every task, two choices.

Every time a task starts, Maestro answers two questions. Those two answers decide how good the work is, how fast it comes back, and what it costs.

  • The Model Router

    Which model should do it?

    Maestro reads the task: what kind of work it is, how hard it is, and what a good result looks like. Then it routes the task to the model that does that work best, at the lowest price. A plan goes to a model that plans well. Code goes to a model that codes well. A short reply goes to a quick, low-cost model.

    The Model Router
  • The Cloud Router

    Which cloud should run it?

    Every model and every app runs on a computer somewhere. Maestro looks at what the job needs, which clouds have room right now, what each one charges, and where the data should stay. Then it puts the job on the cloud that runs it best, and ZB Cloud runs it there.

    ZB Cloud

The Model Router is Maestro choosing the model. The Cloud Router is Maestro choosing the cloud.

What it weighs

Four things, every time.

The best route is the right model for this task, today. Sometimes that is the strongest model there is. Most of the time it is a smaller one that does this one job just as well, for a lot less. Maestro weighs four things every time.

  • Quality

    How good the work will be. Maestro knows which model does each kind of task best, from millions of past results.

  • Price

    What the work will cost. Some models cost a hundred times more than others. Maestro saves the costly ones for the tasks where they make the difference.

  • Speed

    How soon the work comes back. A customer waiting for an answer needs it in seconds. A report for tomorrow can take longer and cost less.

  • Place

    Where the work should run. When data has to stay in one country, Maestro routes only to the models and clouds that keep it there.

Developers can add their own rules on top: always use one model for pictures, spend up to a set amount on each task, keep the data in the US and Canada. Rules for developers

Follow one task

One task, start to finish.

One task from our example store, Northwick Candle Co.: write the code for an email that reminds a shopper about the candles still in their cart. The store sells in two currencies, US and Canadian dollars.

  1. Step 1

    Read the task.

    Maestro sees code work: four files, with tests. It needs a model that writes good code. It does not need the most costly one.

  2. Step 2

    Route to the model.

    The code goes to DeepSeek, which writes strong code at a low price. The plan for the files goes to GPT. The tests go to Sonnet.

  3. Step 3

    Put it on the cloud.

    The work runs where it is fastest and cheapest right now, and the preview of the store goes up on ZB Cloud.

  4. Step 4

    Check the work.

    The tests run. One fails: a price shows in the wrong currency. Maestro gives only that one fix to Opus, a stronger model. The test passes, and a second model reviews the change.

  5. Step 5

    Hand it on.

    The email words go to GPT. The finished work goes to a preview link, and every step, model and cost goes in the record. Maestro remembers what worked, for next time.

The same task, one model for every step$1.45
The same task, routed by Maestro, step by step$0.52

Example task. List prices on 10 October 2026, before the 5% fee. Hosting is separate. Every step and its cost is on the Model Router page.

Smarter every day

Every task teaches it something.

After every task, Maestro writes down what happened. Which plan it used. Which model did the work. How good it was. How fast. What it cost. How many tries it took.

Every one of those notes makes the next route better. More tasks, better routes. Better routes, better price, better speed, fewer tries. Better price and speed, more tasks.

  • When

    A new model arrives.

    Maestro tries it on real tasks and learns what it is good at. The day it becomes the best choice for a kind of work, Maestro starts using it. Your stack gets the new model with nothing to change.

  • When

    A price drops.

    The next route uses the new price. When a good model gets cheaper, more of your work goes to it, and the same work costs less.

  • When

    A cloud fills up.

    Maestro sends new work to the next best cloud that keeps the same rules, so the work keeps moving at the same speed.

The stack you use next year is smarter than the one you use today, and you did nothing to make it so.

Who sends it tasks

Your AI team uses Maestro. So can your agents and apps.

  • AI teams

    Every task your AI team does.

    Lead, Product, Design, Coding, Marketing, Sales, Operations and Service. Each task they start goes through Maestro, so each one gets the right model and the right cloud.

    Meet the AI teams
  • Developers

    Every model through one door.

    Building an agent or an app? Send Maestro its tasks. It routes each one to the right model and the right cloud, checks the work, and follows your rules. One balance for every model.

    The Model Router
  • Governments

    Sovereign AI, with the best of every model.

    Sensitive work stays inside your border. Everything else runs on the best cloud for the job. Maestro sends every task to the right one.

    The sovereign AI offering

Proof

A whole store, routed by Maestro.

Northwick Candle Co., built from start to finish by an AI team: the storefront, the shopping helper and the eighteen systems behind them. 3,559 AI tasks across 12 models. Seventy-nine tasks failed a check and went to a stronger model; those retries are inside the price.

Fable, one model for everything$720.00
ZB Stack, routed by Maestro, with the 5% fee$29.10
96% less than one model for everything

Example build, not a customer result. One model = Claude Fable 5.1 at list price on 10 October 2026. Image tasks priced on Gemini in both columns. The $29.10 total includes the 5% fee. Costs shown before sponsorship. Hosting is separate.

Maestro

Give Maestro a task. Get the best of every model.

Billions of agents. Millions of models. One Maestro.

Bring an idea.

Apply for access

How it works →