OSM Agent — Managed State Classification via the Agents API

The OSM agent is the fully-managed counterpart to the client-side embed-and-KNN pattern in atai-newton-omega-model: instead of fanning out /query embedding calls and classifying locally, you hand the platform a CSV and the platform runs the whole graph server-side:

source → interpolate → window → windowInterpolate → samplingRate → limitValues → encoder → classifier → sink

You don't build or host anything: the platform ships canonical "OSM Quick Start" bundles with the six-state Volve classifier and its windowing already pinned. One run = one agent instance = one input file. You upload the CSV, resolve the pre-packaged bundle by name, run it, poll until terminal, and download one output CSV of per-window predictions.

When to Apply

• Classify operational states over a full CSV of sensor records with no client-side ML and no classifier of your own — the maintained bundle embeds and classifies every window
• Demo or evaluate the managed OSM path as a deployed, repeatable batch job
• Score the managed predictions on held-out slices against a ground-truth sidecar