Really like this approach! The fact that everything runs in the browser and nothing is stored makes it feel both fast and secure. One thing I’m curious about: how well does the AI handle edge cases in larger, messy OpenAPI specs? I imagine a lot of teams will throw pretty complex files at it.