The model does not respond
Check which provider and model the chat is using. Distinguish a missing credential, expired account access, unavailable local model, network problem, and provider rejection. The error should guide the next check.
Confirm that the selected model and connection are supported by the version you installed. A model being listed does not establish all of its capabilities.
You reached a provider limit
Your provider controls its balance, subscription entitlement, and request limits. Check that arrangement before retrying repeatedly. Provider-funded free models can have capacity and eligibility limits.
If you deliberately select another route, check its cost and capabilities. Do not assume that signing in again creates an inference allowance.
Earlier data is missing
The Python runtime may have restarted or been lost. Ask the model to check what is available now, then look for saved output files or supported checkpoint values.
Checkpointing and restoration are optional and off by default. Saved output files, supported restored values, and approved memory have separate lifetimes. See State & continuity.
A browser or application task is waiting
Check for first-use setup, a browser/profile choice, or an application prompt. In personal browser mode, a locked source profile may require you to close that browser and retry. VARIANT-1 does not close it automatically.
Before repeating a write, click, send, or navigation, establish whether the earlier action already happened. A missing follow-up observation does not always mean the action failed.
A task was interrupted
Inspect what completed, what stopped, and what may remain active. Stop does not undo external changes. If an action’s outcome is unclear, inspect the destination before trying again.
Give the next turn the specific correction or missing result rather than assuming a full restart is necessary.
Report an issue clearly
- The exact app version and Windows environment.
- The selected provider/model, without credentials.
- What you expected and what actually happened.
- A small set of reproducible steps using non-sensitive samples.
- The visible error and any relevant, reviewed evidence.
- Whether the issue blocked the task or left an uncertain external effect.
Report reproducible source-preview bugs through GitHub Issues using redacted examples. For sensitive reports, follow the repository’s security policy. Check Release status for installer availability.