Skip to main content

Conversation Call Logs

Open Luna's call log panel for the current conversation. Select a run, inspect its event sequence, and open an event to read the detailed payload. Load more runs/events when the panel indicates more results; the first page is not the complete history.

For a failed action, locate its run and the command event, then inspect the returned error code and details. A submitted operation or tool request is not proof of completion: check its result and any background operation status. Include the run ID and event sequence when reporting a problem. Detail payloads may be truncated; do not interpret missing text as a successful result.

These logs describe conversation execution. Resource Management model usage logs describe model calls and accounting; they answer different questions.

The page reads tenant Runtime routes under /api/soln/luna/threads/:threadId/call-logs. Run pagination uses beforeRunId; event pagination uses afterSequence; a single event is identified by run ID and sequence. Access is checked against the authenticated conversation scope. This is a page capability, not a new CLI command.

Read diagnostic events​

  • Command discovery records command selection; it does not mean a command executed.
  • Command input rejected identifies schema validation failure. Read its diagnostic fields before correcting inputs.
  • Tool input recovery records a recovery attempt, not successful business execution; check the following tool result.
  • Auxiliary model calls are separate from the main response. Use their start/completion events to distinguish model work from command execution.
  • Context compaction and prompt/token budget events describe context processing. A completed compaction does not mean the requested task is complete.
  • Model output incomplete requires checking the recorded reason and any subsequent recovery/result before claiming completion.

Read the actual event detail and sequence rather than inferring success from an event name. Available payload fields depend on the event; absent or truncated fields are not evidence of a successful command.