Skip to content
HearthworkHearthwork

Diagnose

hearthwork diagnose reads the numbers the host records for every generation request. It does this passively: streams are never buffered or delayed.

What is recorded: client, API, outcome, queue wait, time to first byte, duration, prompt, cached and generated tokens, speeds, and the request’s size against the context. The numbers go to logs/requests.jsonl in the data folder (rotated at about 5 MB). No prompt or response text is recorded, and nothing leaves your computer.

Terminal window
hearthwork diagnose                      # everything recorded
hearthwork diagnose --since 30m --client opencode
hearthwork diagnose --last 50 --json
hearthwork diagnose --export report.json # numbers only: no prompts, code, credentials, paths or device names

How to read the report

Section titled “How to read the report”

The report separates two kinds of findings:

  • Observed findings are backed by the numbers, for example time spent waiting in the queue because several clients shared the slots.
  • Possible causes are explanations the numbers are consistent with. They are labelled as such.

Each finding comes with a suggestion. What the backend does not report is shown as “unavailable”. Model load time, memory use and time in agent tools are not measured.

Opt-in raw capture

Section titled “Opt-in raw capture”

For debugging, set HEARTHWORK_CAPTURE=1 before starting the host to also write full request and response bodies (your code and conversations) to logs/capture.jsonl. It is a local file, never exported, and off by default. Delete it when you are done.