Built a prototype to classify API deprecations — it works, now turning it into a product
built a script to test whether LLM classification of API changelog entries was feasible. Specifically: can you reliably distinguish "breaking change" from "additive noise" across inconsistent vendor formats?
Turns out yes, well enough to build a product: Sunsetr.
Three things I've learned so far:
The Moat is Classification: Anyone can scrape an RSS feed. Reliably extracting "this endpoint is sunset on X date, replaced by Y" from unstructured text is the hard part.
The Pitch: "Uptime monitors tell you when an API is down; Sunsetr tells you when it's about to change forever."
The Gap: Releasebot ($59/mo) validates that people pay for aggregation, but they don't do specific deprecation detection.
Landing page + waitlist at sunsetr.nohall.dev. Targeting $19/mo at launch. Happy to talk through the technical approach if anyone's building something similar!
