Three weeks after a team wired a free model allowance into their demo, the compliance review landed. Question one: where does prompt data go? Nobody had an answer. Question two: what happens when the quota runs out mid-sprint? Silence. The demo worked. The architecture didn't.
I've watched this pattern repeat on more than one team. A free tier looks like a gift until it becomes a dependency. Then it's a trade you never consciously agreed to. So let's make the trade explicit before you sign it.
This is a decision framework for one recurring question: should you build on a free managed AI dev platform, or self-host the stack yourself? I'll use MonkeyCode as the concrete example. It's an open-source AI development platform that currently offers a free server option plus free model access — including a 10M-token allowance at the time of writing. Disclosure: This article was prepared as part of MonkeyCode's product outreach.
The framework has five gates. Each gate is a question you answer with evidence, not vibes. If you can't answer one, you haven't decided yet. You've guessed.
Gate 1: Where does the data boundary sit?
