Documentation Index
Fetch the complete documentation index at: https://mintlify.com/withastro/flue/llms.txt
Use this file to discover all available pages before exploring further.
flue logs lets you observe what an agent did during a run. It can tail a live run in real time or replay the persisted event history for a completed run. It is read-only and never invokes the agent.
Usage
Getting the run ID
The run ID is available in two places:- The
x-flue-run-idresponse header on any POST to/agents/<name>/<id> - The
run_startevent in the SSE stream:{ "type": "run_start", "runId": "..." }
flue run, the run ID is also printed to stderr: [flue] Run ID: <runId>.
Flags
Run ID to observe. Positional argument. Obtain it from the
x-flue-run-id response header or from the run_start event.Base URL of the running Flue server. Defaults to
http://127.0.0.1:3583 — the default address for flue dev.Keep streaming live events after any buffered history has been replayed. Mutually exclusive with
--no-follow. By default (neither flag set), flue logs automatically follows if the run is still active and exits if it is already complete.Replay persisted events and exit immediately. Does not open a live stream.
Event index to start streaming from. Treated as
Last-Event-ID on the SSE stream. Use this to resume from a known offset without replaying earlier events.Comma-separated list of event types to include. Events of other types are silently discarded. Example:
tool_call,log,run_end.Maximum number of events to emit before exiting. Applies to both one-shot and streaming modes.
Output format. Accepted values:
pretty(default) — Human-readable output to stderr, matching the format used byflue run.json— One JSON object per line, written to stdout.ndjson— Newline-delimited JSON, written to stdout. Identical tojsonin current behavior.
Auto-follow mode
When neither--follow nor --no-follow is set, flue logs queries the server for the run’s current status:
- If the run is active, it opens a live SSE stream and follows until the run ends or you press Ctrl+C.
- If the run is completed or errored, it replays the persisted event history and exits.
Event types
The following event types may appear in the stream:| Type | Description |
|---|---|
run_start | Run has begun. Contains runId and agentName. |
run_end | Run has completed. Contains durationMs and result (or error). |
text_delta | Incremental text output from the model. |
thinking_start | Model has begun a thinking block. |
thinking_delta | Incremental thinking content. |
thinking_end | Thinking block complete. |
tool_start | Tool invocation has started. Contains toolName and args. |
tool_call | Tool invocation has finished. Contains result or error. |
turn | One LLM round-trip has completed. |
operation_start | A prompt, skill, or task call has started. |
operation | A prompt, skill, or task call has completed. |
compaction_start | Context compaction has started. Contains reason and estimatedTokens. |
compaction | Context compaction complete. Contains message count before and after. |
log | Structured log message from the agent runtime. Contains level and message. |
idle | Agent is waiting (between turns or operations). |
Exit codes
| Code | Meaning |
|---|---|
0 | Success |
2 | The run ended with an error (run_end.isError === true) |
130 | Interrupted by Ctrl+C |