Desktop troubleshooting
For KeenKind API requests from Codex, ZCode, or another tool, start with API troubleshooting.
First confirm the current chat, directory, model, and permissions. Many failures come from a mismatch between the task context and the current page.
Model does not respond
Jump to Model does not respond- Check that the configuration and model are enabled in Settings > Configuration.
- Check endpoint, model ID, and key.
- Match the provider and API type.
- Send a simple message in a new chat to rule out current-chat state.
- Check provider rate-limit, balance, or region errors.
Choose the matching provider for services such as DeepSeek even when they support OpenAI Chat Completions.
Tools do not run
Jump to Tools do not run- Check the permission mode near the composer.
- Confirm the working directory.
- Check required Skills, MCP, or plugins in Settings > Extensions.
- Expand the failed call for its specific error.
- Stop the current task before retrying to avoid overlapping operations.
Files or images cannot be used
Jump to Files or images cannot be used- Check format, size, and count limits.
- Confirm files are still in their original locations.
- Resolve workspace-relative paths from the current project.
- Try previewing Office documents, PDFs, or images in the right panel.
- Password-protected, damaged, or scanned files may lack extractable text.
Browser shows the wrong page
Jump to Browser shows the wrong page- Browser state belongs to each chat; check the current chat.
- Inspect the selected tab and URL.
- Close unused tabs to avoid repeated page openings.
- Complete sign-in, CAPTCHAs, QR codes, and payments yourself.
- After a page changes, let KeenKind inspect it again before continuing.
Extension does not load
Jump to Extension does not load- Disable and re-enable it.
- Check the platform, command path, and environment variables.
- Inspect details or plugin debugging diagnostics.
- Update or reinstall the extension.
- For persistent third-party failures, contact its author with version and error details.
Subagents and persistent tasks
Jump to Subagents and persistent tasks- No tool execution: check the role’s approval policy and allowed risk categories.
- Stuck in queue: concurrency slots are occupied; the run starts when one becomes available.
- Role edits have no effect: existing runs keep their original snapshot; create a new run.
- Long task stops progressing: check whether information or plan confirmation is needed.
- Planner or verifier unavailable: check for usable models in Settings > Persistent tasks.
Collect logs
Jump to Collect logsProvide support with:
- KeenKind version.
- Operating system version.
- Model provider and API type.
- Reproduction steps.
- Error screenshots.
- Logs for the relevant time range.
Windows production logs are usually in:
%LOCALAPPDATA%\Keenkind\production\logsDevelopment logs are in %LOCALAPPDATA%\Keenkind\development\logs.
Remove API keys, cookies, personal file content, and other sensitive information before sending logs or screenshots.