# AI sync & index status

So that AI chat and semantic search can find your content by meaning, Lalabase turns notes,
tickets, posts, documents, meetings, conversation analyses, documentation pages and the file
attachments on tickets and comments into a searchable vector index (so-called **embeddings**). This
page is a status overview only: it shows how complete that index is — for the whole
organisation and per project. Normally there is nothing to do here.

## What the AI index is for

An embedding is a mathematical representation of a text. Only what is embedded can the AI find
by **meaning**, not just by keyword. If a piece of content isn't embedded yet, it simply won't
appear in AI answers or semantic search.

## What the overview shows

Right at the top there is a plain-language summary: whether everything is processed, how much
is outstanding, whether anything failed, and next to it the **Reprocess** and **Refresh status**
buttons. Normally that one paragraph is all you need; everything below it is detail.

Below it you see the organisation totals and a per-project breakdown with a progress
ring. The project table lists every content type Lalabase prepares: **notes**, **posts**, **tickets**,
**documents**, **documentation pages**, **meetings**, **conversation analyses**, **ticket
attachments** and **comment attachments**. Tickets are always embedded. Notes, posts,
documents and meetings only when AI preparation is enabled for them. Conversation analyses
only when they are published and released for search. Attachments only when text can be read
from the file, and never on a private comment.

<div class="docs-callout docs-callout--info">
  <div class="docs-callout__title">Above and below count the same</div>
  <p>The summary and the project table read the same list of content types. What the summary names as outstanding can be found in one of the projects below. Content whose preparation failed counts as an error in both places, not as outstanding. <strong>Reprocess</strong> still tries it again.</p>
</div>

### The vector space named in the summary

The summary also names the **active vector space** (provider, model and dimensions), the one AI
chat and search currently read from. That is not decoration: an organisation can hold several
spaces.

The figures above it are **space-independent**, though: they say how much content is still
waiting or has failed, not which space the finished content sits in. During a model switch the
page can therefore read "everything processed" while the fresh content lives in the new space
and search still reads the old one. The notice about the running switch sits right next to it.

If it says the space cannot be determined, the AI access is missing or blocked. A catch-up run
will not help then; check **Access** in the **AI** area.

### While a model switch is running

When the embedding model is switched, Lalabase builds the new vector space in the background.
The summary says so explicitly and shows the progress. Content counted as outstanding is normal
while this runs. The build continues on its own, and clicking **Reprocess** does not speed it
up.

## Processed, pending, errors

- **Processed** — the content is in the index and findable by the AI.
- **Pending** — the content is queued for preparation but not processed yet.
- **Errors** — preparation failed, for example because of a provider problem.

## Reprocessing pending content

**Reprocess** kicks off preparation for all still-pending content. It's a catch-up run: already
processed content is left untouched, nothing is indexed twice. Processing runs in the
background — the numbers update after a short while, and **Refresh status** loads the current
state. The button sits in the summary at the top and only appears when there is genuinely
something to catch up.

<div class="docs-callout docs-callout--info">
  <div class="docs-callout__title">Why does nothing happen?</div>
  <p>Reprocessing needs AI to be enabled. If AI is switched off for the organisation or the AI quota is used up, nothing is processed — enable AI first in the <strong>AI</strong> area under <strong>Access</strong>.</p>
  <p>Background processing itself may also be down. The summary then says so outright instead of continuing to report "processing". A catch-up run does not help in that case: the content is queued and stays queued until processing runs again.</p>
</div>

Only account administrators can see the index status and trigger reprocessing.