Start with the route that ran the generation: local Codex or a hosted model.
First checks
- Confirm the Mac is online.
- Reopen the Project and read the node’s current status.
- Check the selected model and settings.
- Confirm the account has enough storage for another result.
For local Codex, open Settings > Codex and confirm Local Codex is ready. For a hosted model, check the available credit balance and the price shown on Generate.
Local Codex failed
Follow Codex isn’t detected when Leafage reports a missing install, signed-out account, old version, or invalid custom path.
OpenAI may also reject a request because the ChatGPT plan has reached a usage limit. Leafage cannot change that limit. Wait for Codex availability to return or select a hosted model and pay with Leafage credits.
Hosted generation needs more credits
Leafage blocks submission when the available balance cannot cover the displayed maximum charge. Buy a top-up or plan, wait for the purchase to confirm, then refresh the balance.
Monthly credits are spent before non-expiring top-ups.
A hosted job failed
Leafage reserves the maximum charge while the job runs. A confirmed failure, refusal, timeout, or successful cancellation releases the reservation when no usable output was delivered. When the provider outcome is still unknown, the credits remain reserved while Leafage reconciles the job.
A provider result with an unknown status can remain pending for up to one hour while Leafage reconciles it. Do not submit the same prompt again unless you want a separate job and possible charge.
Storage limit reached
Existing Projects remain readable and exportable, but Leafage blocks jobs that need to store a new result. Delete unneeded work or move to a plan with more storage.
The app closed during a hosted job
Reopen Leafage and the Project. Hosted jobs use a server-side job record, so Leafage should reconnect to the original run instead of creating a second one.
Contact support
Include your account email, Leafage version, Project name, node type, selected model, approximate submission time, exact error, and whether the balance shows reserved or charged credits. Remove sensitive prompt content that support does not need.