What is pinwheel?

Pinwheel is a small server, written in the Go programming language, that sits between an app and the AI companies it talks to, such as OpenAI, Anthropic, OpenRouter, Groq, and Gemini. Instead of giving every user or app your real provider API keys directly, you run Pinwheel with your real keys stored safely on your own server, and it hands out its own local keys to whoever needs access. Each local key Pinwheel creates can be limited in specific ways: you can restrict which AI models it is allowed to use, set a maximum amount of money it is allowed to spend, and give it an expiration date after which it stops working. This is useful if you want to let other people, teammates, or separate parts of an app use AI features without giving them your actual, unrestricted provider keys. Setting it up involves creating a text file with your real provider keys, then using a built in command line tool with a simple interactive menu to generate a new limited key, choosing its name, which models it can use, and any spending or time limits. Once created, that key works with a running proxy server that listens for requests, checks the key against a small local database to confirm it is valid and within its limits, forwards the request to the correct AI provider, and streams the response back to whoever made the request, the same way a normal API call to that provider would work. The project requires at least one AI provider API key and a recent version of Go installed to run.