What the agent does today, why the set is the same in every channel, and what it refuses.
The set is one
The agent's operations are a subset of the API's, by contract
Nothing exists in the agent without existing in the API, so your team can automate everything a person does by talking.
And nothing the agent does escapes the record the API keeps: the same authorization, the same metering, the same lineage.
What it does
Read, measure, compare, translate, with the source
It opens each document in its format’s unit (your documents), answers from what the engine measured, with the address, compares two versions with a count, and walks the neural network of your vault link by link. When a question becomes a request, it triggers the operation: translate is the first.
What it refuses
When it is not in your documents, it is built to say so
When something is not in your documents, the agent is built to say so instead of filling the gap. It does not have your file to guess from; it has what the engine recorded, and the record either holds the answer or it does not. It is a language model: the address it gives you is how you check it. Why: the conversation is with the AI.
The limit. The agent reads what was extracted on entry, never the file. What was not extracted does not exist for it. The text sent to the model follows the provider's retention terms. This page and the privacy page say so; in the vault, the agent's first line does too.
How a capability lands on this page
Three questions before a sentence is published
Is there a written decision, was it measured, and can anyone reproduce the measurement. What passes all three is a receipt. The rest stays declared as practice, with no present-tense verb and no number. Where coverage ends is in trust; the platform underneath is in platform.
Where the set reaches you: interfaces.