Capabilities
An AI Capability is one reusable unit of intelligence inside a solution — one specific prediction or generation, with a clear contract. An Asset Intelligence solution might contain three: Predictive Maintenance, Remaining Useful Life, and Failure Prediction.
Capability types
Every capability has a type, which describes the kind of question it answers.
Classic machine learning:
| Type | Answers |
|---|---|
classification | Which category does this belong to? |
regression | What number should we expect? |
forecasting | What will happen over the coming periods? |
clustering | Which natural groups exist in this data? |
recommendation | What should we suggest next? |
anomaly_detection | Is this unusual? |
time_series_anomaly | Is this unusual over time? |
Generative AI (built in AI Studio):
| Type | Produces |
|---|---|
text_generation | Free-form text from a prompt. |
summarization | A summary of the given text. |
extraction | Structured fields pulled out of text. |
embedding | A numeric representation for search and retrieval. |
Agentic AI: the agent type runs a goal-driven loop that can call tools —
see Agents.
All types share the same home (a solution), the same quality gate, and the same serving endpoint shape — a consumer of the API cannot tell a classic model from a generative one.
The contract
What makes a capability reusable is its contract: a declaration of the inputs it needs (field names and types) and the outputs it produces. Because the contract is explicit:
- Every prediction request is validated before it reaches the model — a missing or malformed field is rejected with a clear reason instead of producing a silently wrong answer.
- Dates and timestamps are normalized, so "2026-07-17" means the same thing at training time and at serving time.
- The platform can check, before a deployment starts, that the model will actually receive every field it was trained to expect.
One active model at a time
A capability is served by exactly one production model version at any moment. Improving the capability means training a better model and promoting it; the switch-over is instant, and rolling back is the same pointer switch in reverse. Your applications never notice — the endpoint stays the same.
Working with capabilities in the Studio
- Open a solution and add a capability: choose the type, name it, and declare the input/output contract.
- Set its acceptance criteria — the bar every candidate model must clear.
- The capability detail page is mission control: contract, training runs, evaluation results, the active model, and (for the automated path) the retraining policy all live there.
- A capability moves through Draft → Configured → Trained → Deployed → Monitored → Retired as you work; the status badge on the list tells you where each one stands.
Next
- Set the quality bar: Acceptance criteria.
- Feed it: Datasets.