AI Coach

AI Coach is an optional Pro assistant that helps you describe packs and feed preferences. It proposes changes; it does not silently rewrite settings or control the live timeline.

• Local feed filtering never waits on AI.
• Cloud processing is off by default and requires consent.
• You always review a proposal and choose which items to apply.
• No OpenRouter or provider API key is stored in the extension.

Before each request you can expand What will be sent. Only the fields for that single action are included (for example a short description, or one pack’s metadata and terms). Feed history is not bulk-uploaded.

Requests go to the XpressPurge API, which routes them through OpenRouter to a Gemma model on an eligible upstream provider. Free and paid routes may differ in capacity; a route or data-policy change asks for consent again.

• Pack Studio (Options → Filters) — describe a new pack, or improve the selected custom pack.
• Filters — AI Coach on the Smart topics card helps you refine what to keep seeing or see less of.
• Modes — Describe the feed you want / full setup feed mode (Smart topics, kits, optional exact rules, strictness). Approved Smart topics replace the mode’s list rather than only merging forever.
• Activity → Recently removed → Wrong call — after a local restore, optional Help prevent this offers deterministic pack adjustments first; AI Coach is a second, optional step.

Separate from the cloud AI Coach, Coach can show occasional one-tap suggestions right in your feed — for example a standing rule for an author you keep hiding, or a topic kit that is over-hiding. These suggestions:

• are computed entirely on this device from your local hidden log, corrections, and learned examples — no network, nothing leaves the browser;
• are never applied without your tap, and dismissing one tells Coach not to repeat it;
• can be turned off at Options → Settings → Filter suggestions.

AI Coach needs a live Pro session on this browser (entitlementToken from Activate or Restore). A local DEV_PRO paint is not enough.

1. Review the summary and itemized list.
2. Uncheck anything you do not want.
3. Click Apply selected.
4. Use Pack Studio Undo when a pack revision was saved.

If AI is unavailable, rate-limited, or you decline consent, every local feature continues to work.