RadarModel Launches, Watching the AI Models Under Your App

RadarModel, the studio’s twelfth project, watches the models an app depends on. Its first product, RadarModel AI, reads the model lists of OpenAI, Anthropic, Google and OpenRouter, compares each with the last scan and says when a model you call is gone, retiring, or has a new limit or price. Version 0.1 came out on 17 October, followed the same day by 0.1.1, a small wording fix in the report. It’s one Go binary for Linux, macOS and Windows, open source under Apache 2.0.

You list the models your app calls and only those can raise anything. A watched model gone or a context limit cut is breaking, a price rise or a retirement date is a warning, and the forty new models a provider added are info. Reports go to the terminal, a webhook, Slack or email, and check exits 1 on a breaking change, so it works as a CI gate.

Most of the testing went into not crying wolf. The test write-up breaks a provider’s answer in 14 ways, from a 500 to a cut-off body to a half-loaded page, each run three times in one recorded run, and none is reported as a removed model. A removal only counts as breaking once a second scan agrees, which covers lists that come back short with no error. A live scan of OpenRouter’s public list read 470 models.

The limit is what providers publish: OpenAI’s list has little beyond model IDs, so there it can only report a model gone. Next on the same engine is RadarModel Schema, which points the same loop at Postgres, MySQL and SQLite for changes made outside migrations. Where RollupDB is a database the studio built, RadarModel keeps watch on models other people run for you.