Troubleshoot the AI Assistant
Identify whether the problem is the request, generation, Preview, data, or access before making another broad change.
The assistant uses the wrong source
Selected resource and exact table names
Select the intended source and clarify the request.
A field is missing
Source schema and connection access
Inspect the connection before changing the UI.
The wrong page or component changes
Target and scope in the prompt
Request one focused correction.
A file will not attach
File picker restrictions and upload error
Check the attachment workflow.
Dictation is unavailable
Browser permission and microphone input
Check dictation setup or type the request.
Preview is blank or disconnected
Session status and recent logs
Inspect Console; use Restart Preview when appropriate.
An action fails
Inputs, source permissions, and actual source result
Test once with a safe record.
A user sees unexpected records
Authentication, user properties, and record rules
Compare user experiences and check enforcement.
Usage is limited
Account allowance and usage messages
Check How credits work.
Preview limit reached
Open an unused app preview, select … → Stop Preview, and return to the app you want to work on. See Control app Preview.
The assistant appears idle during a tool call
Long-running actions can show progress while the assistant uses tools. Check the active action and any available details before sending the same request again. If progress stops or an error appears, record the action, time, and error for support.
Preview is blank during generation
Wait for initial app generation to finish. The builder holds Preview until that step completes. If generation is complete and Preview remains blank, check Console and logs.
Recurring request timeouts
A fix for recurring request timeouts has been deployed and is being monitored. If the error continues, record the time, prompt, and visible error for support. Before retrying a request that writes data, check whether any records were already created or changed.
Ask for help with a reproducible case
Include the app page, time, relevant prompt, selected resource and model, expected result, actual result, and a sanitized error. State whether the problem also occurs in the published app.
Before repeating a create, update, or send action, confirm whether the first attempt already completed. Keep a failing test case so you can verify the eventual fix.
Was this helpful?