Asking the assistant

Context, footnotes, the general-knowledge label, answer depth, and what an answer's details tell you.

What it is for

The assistant answers questions from the documents you can reach, with the source named every time. It does not browse, and it does not answer from documents you were not granted. When your documentation does not cover a question, it says so, and if it goes on to answer from general knowledge the answer is marked as not from your documentation.

The three things most people do

Ask in plain words. Type the question and press Enter (Shift and Enter adds a line). The answer carries numbered footnotes; each opens the source at the passage it came from, in a new tab. The chips under the answer name the documents it drew on and open them in Sources. Names of tables and dictionary entries in the answer are links to their entries.

Choose a context. All sources is everything you can reach. Choosing a collection narrows the answer to that project or subject, which is the right move when two collections could both answer and you want one of them. The answer card says which context it was asked in.

Open the details. The i icon on an answer shows where it came from, which model wrote it, the depth used, how many passages were read, the tokens it cost, and how long it took, with a button to ask the same question again one level deeper.

Depth

Answer options, next to the question box, sets how much of your documentation one question may read. Standard reads up to fourteen passages and suits nearly every question. Extended reads about twice as much, for a question that spans a whole document. Maximum reads up to sixty passages and whole documents, for audits and full listings. Which depths are open depends on your plan on the included model (the panel says which, and why a setting is grayed); a workspace that answers on its own model key chooses any depth.

The one mistake to avoid

Reading an answer marked as general knowledge as if it came from your documents. The label is there for that reason; the footnotes are the test. An answer with footnotes came from your files; one without came from the model.