Sidebar → AI Cache (the "Management" group) — view and manage the AI answer cache. This section is available to the Admin and Owner roles.

#What the cache is and why it's there

When AI produces an answer to a query, LiKE saves it. If the same (or one reduced to the same canonical form) question is asked again, the answer is served from cache: instantly and without spending tokens. In Analytics, such answers are marked as CACHE_HIT.

The cache self-cleans by meaning: when the source articles of an answer change, the corresponding entries are automatically marked invalid, and on the next query the answer is regenerated. This screen is for viewing the cache contents and managing them manually if needed.

#Metrics (at the top)

Metric What it means
Valid How many entries are currently fit to be served (out of the total)
Invalid How many entries are marked stale (not served, regenerated on request)
Uses How many times answers were served from cache (saving tokens and time)

#Entries list

A table of cached answers. Columns: Question, Answer (preview), Language, Type (answer type), Sources, Uses, Status (valid / invalid), Date. Clicking a row opens the details: the full question and answer, language, type, number of uses, status, and Canonical ID.

Filters: search by text, language, status (valid / invalid). The list is paginated.

#Actions

Action What it does
Invalidate cache Marks entries invalid — by Canonical ID (and language) or all at once. The entries aren't deleted but stop being served; on the next query the answer is regenerated
Clear old Deletes entries older than the specified number of days (frees up space)
Delete entry (the trash icon in the row) Deletes a specific cache entry

When manual control comes in handy. Usually the cache updates itself when articles are edited. Manual invalidation is useful after a bulk content update — to guarantee fresh answers are served. Full cleanup of the entire cache in one action is in Settings → Danger zone → "Clear AI answer cache".

#Related sections