Logs and token usage¶

The Logs tab¶

The GeoAgent Logs tab shows what happens behind each reply: the tasks a request was split into, the algorithm chosen for each, the parameters used, any corrections or retries, and the outputs. It’s the first place to look when a result isn’t what you expected.

The Logs tab after a processing request, ending with its token usage

Lines you’ll often see in Processing mode:

Line	Meaning
`WORKFLOW decompose -> 3 task(s)`	How the request was split.
`GEO select: native:buffer (...)`	The algorithm picked for a task, with the reason.
`GEO gather fix: INPUT: 'River' -> layer 'rivers'`	A slip that was corrected before running.
`GEO gather: {...}`	The parameters the algorithm ran with.
`GEO pre-flight check failed: ...`	QGIS rejected a parameter before running; GeoAgent retries with a fix.
`GEO retry plan: bad_parameter -> re-gather parameters for native:buffer`	How a failed task will be retried.
`GEO execute OK -> outputs: [...]`	The task ran; these layers were added.

The same log is written to geo_agent.log in the GeoAgent folder of your QGIS user profile (open it with Settings ▸ User Profiles ▸ Open Active Profile Folder), so it’s still there after a restart.

Token usage¶

Every request ends its log with the tokens it used, summed over all the LLM calls it made, plus a running total for the QGIS session:

Tokens used: 15,408 (input 14,640, output 768) in 8 LLM calls · total this QGIS session: 17,334

Input tokens are what GeoAgent sent: your request, layer information, algorithm details. Output tokens are what the model wrote back. Cloud providers price the two differently; check your provider’s pricing to turn tokens into cost.
A processing request makes several LLM calls: splitting the request, choosing an algorithm and filling in parameters per task, analysing failures, and writing the summary. Multi-step requests therefore use more tokens than a single question.
The counts come from your provider. If a provider doesn’t report them, the line says not reported by the provider.
With Ollama, tokens cost nothing, but the count still shows how much work a request took.

Saving a conversation¶

Export Chat in the Chat tab saves the conversation as a text file.