Reading an answer

An answer has four parts, from top to bottom.

The answer

The answer streams in as it is written, with a cursor at the end. Under it, a small line names the model that wrote it. For on-device answers it begins with an on-device tag. You can select and copy the text.

The answer is written only from the passages the search found. If they do not support an answer, the model says so.

The caption

A line of figures sits under the answer, for example:

📊 12 passages · 2 KGs queried · search 41 ms · synthesis 3,200 ms · 5 of 12 in context

Part	Meaning
passages	How many passages the search returned.
KGs queried	How many indexes it searched: the books, the diaries, or both.
search	How long the search took.
synthesis	How long the answer took to write.
N of M in context	How many of the passages fit in the model's reading window. The rest were found but did not reach the model.

Source passages

Source passages lists everything the search found, best first. It is folded away under an answer, and open when there is no answer to read.

Each passage shows the genre, the kind of text, the book and author, the beginning of the text, and a bar for how well it matched. Diary passages show their date. Show full passage expands a long one.

A passage the model actually read carries a green in context tag. When an answer seems to ignore something you know is in the book, look here first: if the passage you expected is not marked, the model never saw it. Try different wording, or narrow the scope.

Render

Under some answers there is a Render button, which asks your own worker server to draw an illustration of the answer. It appears only when a worker is set up and reachable. See Settings and Troubleshooting.

When there is no answer

The app tells you why in the answer area, and the passages stay below, where you can read them. See Troubleshooting.