start with the smallest test
- Confirm Enable Globally is on.
- Confirm the current application is not in Disabled Apps.
- Confirm Accessibility and Input Monitoring say Granted.
- Confirm Apple Intelligence is available or the selected GGUF model says ready.
- Quit competing autocomplete utilities while testing.
nothing appears anywhere
- 1. Check permissions. Open Cotabby Settings → Permissions. Both required rows must show Granted.
- 2. Check the engine. Open Engine & Model and confirm the Apple runtime is available or the local model is loaded.
- 3. Use a normal field. Passwords, passcodes, card fields, and other secure inputs are intentionally ignored.
- 4. Relaunch after permission changes. macOS can require the app process to restart before a new grant takes effect.
- 5. Replay onboarding. Settings → General → Open Welcome Guide walks through the required state again.
Follow the complete permission recovery steps if a row remains stuck.
one app or website stays quiet
Custom browser, Electron, and canvas-based editors expose different Accessibility trees. First confirm the app is not disabled, then test a standard input on another website in the same browser.
- Restart the affected app after granting Accessibility.
- Try its plain-text field before its rich-text or canvas editor.
- In VS Code or Cursor, keep integrated terminal suggestions off unless you intentionally enabled them.
- Check the current compatibility reference.
popup or misplaced ghost text
Automatic display uses inline ghost text when the host reports reliable caret geometry and a popup card when it does not. A popup is therefore sometimes the safe fallback, not a failure.
Open Settings → Appearance → Suggestion Display to try Automatic, Inline, or Popup. If text remains badly positioned, record the app, exact field, macOS version, and which display mode was active.
model download or loading fails
- Confirm enough free disk space remains for the selected file.
- Cancel and retry an interrupted download from Engine & Model.
- Open the models folder and confirm the file ends in
.gguf. - Use Refresh after adding or removing a model outside Cotabby.
- Try tabby-2-nano or mini if a larger model exceeds available memory.
Apple Intelligence requires no GGUF download, but it must be enabled and available for the Mac, OS version, language, and region.
suggestions feel slow or heavy
Larger models consume more memory and can finish after you have already typed past the suggestion. Start with a smaller model before changing every timing control at once.
- Try tabby-2-mini, then nano if memory pressure remains high.
- Enable Fast Mode to skip screenshot/OCR context.
- Disable clipboard or app context temporarily to isolate context work.
- Use Settings → Performance for live CPU/memory and opt-in request timing.
send a useful report
Include Cotabby version, macOS version, Mac model, engine and model, app and version, website or editor, display mode, and the shortest steps that reproduce the problem. Never attach private writing or screenshots containing sensitive information.
Submit it through Cotabby feedback or open a public issue in the GitHub repository.
related help
Grant Cotabby macOS Permissions Safely
What each macOS permission enables, how to grant it, and how to recover a stale grant.
Choose a Cotabby AI Engine and Local Model
Apple Intelligence, curated GGUF models, custom downloads, Hugging Face, and LM Studio.
Getting Started with Cotabby on Mac
Install, grant permissions, choose an engine, and test your first suggestion.