atai-operational-state-monitoring-agent | Classify every operating regime over a sensor CSV, when you have a full labelled library of the states you care about — drilling states, machine modes, process phases. Two paths. Path 1: copy the platform’s demo blueprint (playground-osm-drilling, a classifier and windowing already pinned) into a project, package it, deploy it in batch, run a session over a sensor CSV, and download per-window state predictions with per-class probabilities. Path 2: build an agent on your own labelled data — fit and score windowing and kNN settings with the Optimizations API, promote the best trial to a blueprint, test it once with the Evals API, deliver it through a package and batch sessions, score the delivery against held-back labels, and serve it a window per query (a serving deployment), checked against the delivery. |
atai-operational-state-monitoring-agent-data-prep | Turn labelled sensor recordings into the role files OSM’s Path 2 accepts — an exact sample grid split at gaps, a check that the sensor can tell the states apart, a group-aware split, library-only z-score, one training file per state, continuous scored files, held-back delivery labels — and preflight every platform rule before upload. Local, no API key. |
atai-rare-event-detection-agent | Detect a named rare fault recurring — a pump breakdown, a process excursion — from a handful of labelled examples plus normal data. Nearest-prototype classifier over Omega embeddings, so it tolerates the severe class imbalance of one or two incidents against months of normal operation. Two paths. Path 1: copy the platform’s demo blueprint (playground-red, a pump-breakdown classifier and its windowing pinned) into a project, package it, deploy it in batch, run a session, and download per-window event predictions. Path 2: build an agent on your own labelled incidents — fit one prototype per class and choose the window and distance with the Optimizations API, promote the best trial, test it once with the Evals API, deliver it through a package and batch sessions, score each incident (caught, how soon) and the false alarms, and serve it a window per query (a serving deployment), checked against the delivery. |
atai-rare-event-detection-agent-data-prep | Turn one labelled recording into the role files RED’s Path 2 accepts — an exact sample grid, labels mapped to a background and named event classes, numbered incidents, a check that the incidents and channels can make an agent (incident length against window size, channels that go dark during events, separation that carries over to held-out incidents), build-time-only z-score, one training file per class piece, scored files around each incident with no row shared, held-back delivery labels — and preflight every platform rule before upload. Local, no API key. |
atai-anomaly-discovery-agent | Flag “this no longer looks like normal operation” on an asset with no fault history — fitted on normal-only reference data, so everything it flags is something it was never shown. Per-window anomaly score from a fitted LOF detector with its threshold pinned; covers scoring by lead time and false-alarm rate rather than precision and recall. Two paths. Path 1: copy the platform’s demo blueprint (playground-ad, a bearing detector with its threshold pinned) into a project, package it, deploy it in batch, run a session, and download a per-window anomaly score. Path 2: fit a detector on your own normal-only recordings and choose its settings with the Optimizations API, promote the best trial, test it once with the Evals API, deliver it through a package and batch sessions, score detection, delay and false alarms, and serve it a window per query (a serving deployment), checked against the delivery. |
atai-anomaly-discovery-agent-data-prep | Turn normal-operation recordings (plus a few labelled incidents to score against) into the role files AD’s Path 2 accepts — an exact sample grid split at gaps, a check that the normal-only library is a trustworthy baseline (purity, stability, agreement, enough windows for the detector), library-only z-score, scored files, held-back delivery labels, a calibration file — and preflight every platform rule before upload. Local, no API key. |
atai-label-free-anomaly-discovery-agent | Find what is unusual in a batch of sensor data with no labels and no normal reference — “what in here looks least like the rest, and when?” Runs the od blueprint in batch: copy it into a project with your columns and window, package it, deploy it, and open one session per file. od fits a Local Outlier Factor detector on each file’s own windows (Omega 1.5 embeddings plus per-channel statistics) and returns a score and a label per window; nothing is trained ahead of time. Covers turning scores into reviewable episodes, and grading against maintenance records when they exist. Batch only. |
atai-label-free-anomaly-discovery-agent-data-prep | Prepare files for the Label-Free AD agent — an exact sample grid with gaps kept as gaps, frozen-logger stretches removed, date ranges where anomalies are a small share of each file, and a preflight of every rule od applies (columns, timestamps, enough gap-free windows for LOF). Local, no API key. |
atai-omega-encoder-agent | Get Omega 1.5 embeddings from a deployed agent (the omega-embeddings blueprint) and build your own detector on them. Batch: a blueprint copy per window size, a package, a batch deployment and a session per role. Serving: a serving deployment and a session whose values set the window, one segment per query, checked against batch. Then a client-side kNN detector fitted on normal-only references, with the window, channels, PCA and k chosen on validation, a threshold at a false-alarm rate, one test and a delivery. For direct per-window calls with no deployment, use atai-newton-omega-model. |
atai-omega-encoder-agent-data-prep | Turn recordings of a repeated operation into the role files the Omega Encoder agent embeds — one fixed-length segment per operation, aligned on its onset, split by recording session into library, validation, test and delivery, z-scored from the library only, with labels held back — and preflight them. Local, no API key. |
atai-manual-generation-agent | Turn a recording of a procedure into an ordered, timestamped manual — each step traceable to a frame range and time range in the source video. Copies the canonical mga blueprint (Newton fusion + Whisper, pinned by the blueprint) into a project with your prompt and token budget, then runs it as a batch session per video, or deploys it for serving and sends a video per query. Audio matters: spoken-only content such as safety cautions shows up in the output. |
atai-task-verification-agent | Check whether a recording followed a known procedure — upload the video and a reference SOP, get a per-step PASSED / FAILED / MISSING verdict with a timestamp and a reason. Copies the canonical tva blueprint into a project with your frame count, frame size and token budget, then runs a batch session per recording, or deploys it for serving and sends a recording per query. The SOP travels with each session or query, so one deployment serves every procedure. Read the skill’s reliability section before trusting a verdict: a skipped step whose tool or part is visible on screen can come back PASSED, so PASSED verdicts warrant human review. |