TagChoose help: local AI setup, downloads and saving

Check the extension version first

Open chrome://extensions and find TagChoose. Version 2 introduces clearer AI setup states, cancel and retry controls, full folder paths and safer saving. See release status to check whether it is available in the Store.

Older versions may show less detailed setup information. An extension update cannot make an unsupported device meet Chrome’s model requirements.

AI is unavailable

Use a supported, up-to-date desktop version of Google Chrome. TagChoose requires Chrome 138 or later. Mobile Chrome does not run this desktop extension. Other Chromium browsers may not provide the same built-in model.

Chrome checks your operating system, available disk space, memory or GPU resources, network and policy settings. Google currently documents 22 GB of free disk space for initial setup, and GPU or CPU requirements. These can change; consult the current official hardware requirements.

Use the Check again control in version 2 after updating Chrome or freeing storage. Keep manual saving available rather than changing experimental flags or disabling browser protections.

The model seems stuck downloading

In version 2, select Set up local AI. The popup shows waiting, Chrome-reported download percentage, and model preparation separately. Reaching 100% for download does not mean preparation is finished.

A message that Chrome has not reported progress means exactly that. It does not prove a transfer is active or that setup has failed. Keep the popup open while checking progress, or stop waiting and save manually.

Open chrome://on-device-internals in another tab to inspect Chrome’s model information. Check your connection, available storage and browser policy. Chrome owns the model download; closing the popup stops TagChoose’s current session, and a later request checks Chrome’s current state again.

Suggestions are missing or inaccurate

Use clear folder names and a descriptive bookmark title. The AI only receives the title, URL and folder list, so a vague title or hidden article context can limit relevance. Folder and page languages outside the model’s supported languages may give poor results.

Version 2 rejects malformed model output and unknown folder IDs, uses exact IDs for duplicate folder names, and offers a retry after an inference error. It also leaves manual edits alone if they were made while a suggestion request was running.

Very large folder lists may exceed the local model’s context budget. In that case, use folder search to choose destinations. TagChoose displays the limitation rather than silently dropping folders.

A bookmark will not save

Choose at least one existing, writable folder and enter a title. Open a normal web page rather than a Chrome settings or extension page. If a selected folder was deleted, reopen TagChoose to refresh the list.

Version 2 reports a partial failure if Chrome accepts some destinations but rejects another. It keeps existing bookmarks and lets you inspect the result before retrying.

Get help with a specific failure

Use the feedback portal or email saulius.developer@gmail.com. Include your Chrome version, operating system, extension version, the exact visible message, and whether manual saving works. Share a public example URL only if you are comfortable doing so. Do not send your private bookmark library.