The basics

What is it?

A coding agent you hire with a budget. It quotes the job, books the right NVIDIA Nemotron model for each step on Nebius Token Factory, runs the project's own tests, and hands back a receipt.

Why?

Coding agents spend frontier-model tokens on steps a smaller model could do, and the bill arrives after the work. Here the price comes first.

How do I try it?

Pick a job, set a budget of 1 to 5 cents, and click Hire. The Upgrade job shows every step.

Is it real?

Yes. Every run calls Token Factory (and Tavily, for Upgrade), runs real tests on a copy, and is really billed. A daily cap protects the demo.

What are the limits?

Four preset jobs on two sample projects (the cart and the checkout, in examples/). One job runs at a time; up to 3 visitors wait in line. Nothing you do changes the project.

How do I read the receipt?

Billed is what Token Factory charged. The grey bar is what the same job would have cost if every step had used the model first requested. Saved is the difference.

Source code and README The examples The video
Slash the Evaluator, the mascot

Hire the agent

A coding agent you hire with a budget. It quotes the job, books the right NVIDIA Nemotron model on Nebius Token Factory for each step, and proves the cost with a receipt.

Why: coding agents spend frontier-model tokens on steps a smaller model could do, and the bill arrives after the work. Here the price comes first, and the receipt shows what was billed against what the same job would have cost on the models first asked for.

What happens when you click Hire: Quote → Search (upgrades only) → Plan → Write → Test → Receipt. Plan runs on Nemotron 3 Ultra, Write on the smallest Nemotron that is good enough, and Test runs the project's own tests on a copy.

New here? Watch the 2-minute video.

The project

A small shop: a cart, where total() adds up items, a percentage coupon and shipping, and a checkout, where validateCart() checks orders with Zod 3. Both have tests. The agent works on a copy; nothing is kept.

Pick a job

Loading…

12345

Every run is real and billed by Token Factory. It takes seconds to about a minute and costs a few cents. The jobs are presets on a sample project for now.