Explain this: ask the local model about a failure, verdict, log line or setting #18

Merged

jcoffey-dev merged 2 commits from feature/ai-explain into main 2026-09-26 08:30:46 +00:00

Conversation 0 Commits 2 Files Changed 17

+1089 -48

2 Commits

Author	SHA1	Message	Date
jcoffey-dev	3b85d1f984	Live tracing: read key-values as the stream sends them ... The live stream sends each event's key-values as a JMAP map ({"0": …}), not an array. The live page cast it straight to an array, so Explain threw on every live event, and the timeline never showed an event's details. Live events now go through the same normalizing as stored traces, moved to features/tracing/normalize.ts. Also adds the modification notice to four upstream files this branch changed.	2026-09-26 01:09:49 -07:00
jcoffey-dev	dfb5deeac2	Explain this: ask the local model about a failure, verdict, log line or setting ... An Explain button sits beside four kinds of thing that are hard to read: a queued message's failed recipients (in a small card above the queue form), the Classify verdict on Actions, each Logs row and trace event (stored and live), and each settings field. It opens a panel with the local model's short reading, labeled as the model's and not as fact, with copy, cancel and a clear message when the model is busy, paused or over the hourly limit. The button appears only when the session's aiExplain flag says the server can answer: a superuser, a configured chat model, and explanations switched on. The Local AI page gains an Explanations card for that switch, the model to use, the hourly count per administrator and the longest wait. Needs a server with inbuxa:Explanation/set (sysAiExplain); on older servers the flag is absent and nothing changes.	2026-09-26 00:53:41 -07:00