Read responses and run information
A response can contain its text, reasoning, one or more Tool activities, generated files, a run summary, and memory changes. Each region appears only when the run produced that information.
During generation
After you send a message, the run may move through a thinking state, Processing, and Tool activity. The composer changes Send to Stop generating.
- To end the current run immediately, select Stop generating.
- If a Tool has already started, stopping generation does not undo that Tool.
- If the page appears idle, first check whether a Tool card is still running before retrying.
Reasoning
Use Reasoning above the answer to expand or collapse reasoning that the model chooses to expose. Not every model provides it, and it does not replace the final answer or resulting state.
Tool activity
A Tool call may appear as a separate card. Related calls can also be collapsed into a group or execution process. Expand it to inspect the name, target, input, output, and state.
At minimum, verify:
- The Tool matches the request.
- Parameters, paths, URLs, and target objects are correct.
- The call completed, failed, was denied, or is still running.
- Long output is not hiding relevant information.
- Whether files or external state changed.
A completed Tool card means that call ended. It does not prove the whole task met its goal.
Charts and interactive results
A response can display a bar chart for category comparisons, a pie chart for proportions, or an interactive visual with its own controls. You may see the corresponding actions named barChart, pieChart, or widgetRenderer.
For example: “Compare these three categories in a bar chart and include the values and units in text.” Check the underlying values, units, and data source before relying on the visual. For an interactive result, try its visible controls and verify that the displayed result matches the question.
Displaying a visual does not create a downloadable file. If you need a reusable file, explicitly request the format and verify that the response includes the actual file. A tool result reporting rendered: true is not proof that the visual loaded correctly; if it is blank or broken, ask for the data and explanation in plain text.
Run summary
The summary may show:
- Total duration and generation speed.
- Tool-call count, failure count, or recovered Tools.
- Skills used in the run.
- Verified, Not verified, or Not evaluated.
- Whether memory and activity context were used.
Fields vary by model and run type. Verified means the identified validation has passing evidence. Not verified means validation was in scope but did not produce a passing result. Not evaluated means validation was not part of the run.
Message usage details
On an AI response, open More actions → Usage to see input, output, cache read, cache write, total tokens, time to first token, and generation speed for that message. Fields depend on data returned by the Provider; an absent field does not mean a value of zero.
Use the run summary for a quick task-state overview and message Usage for token and timing detail. They are separate regions.
Memory activity
The memory panel can list memories read during the run and memories that were created or deleted. changed means a referenced memory changed after it was used; it is not a list of memories updated in the current run. Other states can include current, deleted, and unavailable.
If a memory is irrelevant, stale, or inappropriate, stop sensitive work and correct it in Memory management before running the task again.
Files and response actions
File cards can offer preview or download depending on the file type. Copy, regenerate, branch, side conversation, Rollback, report, and delete actions appear according to message type and state. See Message versions, branches, and Rollback and Feedback and privacy.