Help

Debug sessions

Record while you solve a problem, in as many terminals as it takes and with the assistant, and at the end have it written up: what was wrong, why, what fixed it, the exact commands, what to watch out for and how to keep it from happening again. The write-up uses the model you chose for the assistant.

Record

1. Choose Shell › Start Debug Session (⌥⌘R), or the record button in the assistant's panel. The status bar shows that it records, and for how long.
2. Work as you always do. Every terminal's typed lines and output are kept, with the assistant's questions, commands and answers, and what Wayfinder ran for you from a button.
3. Ask the assistant along the way: each question takes along what was typed and run in every tab since the recording started, so it does not suggest what you already tried.

End it

Click the recording in the status bar, or choose Shell › End Debug Session…. Give it a title and, if you like, the symptom in your own words, then:

• Save and Write Up keeps the record and asks the model for the write-up.
• Save keeps it to write up later, from Window › Debug Sessions.
• Discard keeps nothing of it.

Use the write-up

Window › Debug Sessions lists every session, newest first. A write-up has the problem, the cause, the fix, the commands that fixed it in the order they ran with where each ran, caveats and prevention.

• Copy as Markdown for a ticket, a wiki page or a post-mortem.
• Save as Snippet… keeps the commands as a snippet, to run again in a terminal or on a group of hosts the next time.
• Each command has its own copy button, and What was recorded shows everything that was typed, asked and run.

What is kept, and what is sent

A session is a folder in ~/Library/Application Support/Wayfinder/Debug Sessions on this Mac, readable only by you, until you delete it. Nothing of it is sent anywhere until you ask for the write-up. Then the model gets the record with what looks like a password, token or private key taken out, long output cut to its head and tail, and lines that repeat (a progress bar, a log line with its time) sent once.

What it costs

A write-up is one request, not cached (nothing would read the cache again) and set to think little, and the window says how many tokens it took. A long session is sent as some 25,000 tokens at most. In Settings › Assistant, Model for write-ups can name a cheaper model, such as claude-haiku-4-5, for write-ups alone. With Ollama, less of the record is sent, as a local model reads less at once.

Back to Help