Terms

The product

Reinforcement.tech is an alpha compiler. It writes a Python training loop from three inputs: a verifier, a target task, and a base model. It does not train a model in the browser. It is not a chatbot and it does not produce model completions on this site.

No account, no subscription

There is no sign-up, no paid plan, and no auto-renewal. There is nothing to cancel. If you do not accept these terms, close the tab. Tinker billing, if you train, is between you and Thinking Machines Lab.

What you download

The compiled file is yours to run. Running it talks to Tinker, a training API from Thinking Machines Lab. Tinker usage, pricing, data handling, and model access are governed by their terms, not these. You need your own TINKER_API_KEY. This site never writes that key into a download. The environment, DPO, and SDFT files include placeholders. Replace them before you treat a run as a real reward model.

Your responsibility

The software and the compiled scripts are provided as-is, without warranty of any kind. Training can spend money, overwrite checkpoints, or produce a model that is wrong, biased, or unsafe. You are responsible for what you train, what you deploy, and what it does — including harm to people. This site is not a crisis service and does not monitor or refuse self-harm content, because it does not generate conversational output.

Cookbook cards

Cookbook names (Waldo, Context-1, AxiomProver, and the rest) are compiler presets shaped after public writeups on thinkingmachines.ai/tinker. Each card links there. They are not testimonials, quotes, or endorsements by those teams.

Marks

Tinker, Thinking Machines, and the Thinking Machines mark are trademarks of Thinking Machines Lab. Reinforcement.tech is not affiliated with, sponsored by, or endorsed by Thinking Machines Lab. Pixel icons are from the Pixel Icon Library by HackerNoon. This site’s window chrome is a newspaper-on-a-desktop look. It is not an Apple product and does not use Apple’s logo.

These terms can change without notice while the product is alpha. Questions: David Smooke.

Last updated August 31, 2026. See also Privacy.