43.1 Knowledge Sources vs. What the AI Knows

AIUNIFY Chat separates Knowledge management into two related Agent areas:

  • Knowledge sources — where content is added, crawled, uploaded, refreshed, previewed, reindexed, or removed.
  • Knowledge (what AI sees) — where you inspect the Documents and Chunks actually available to the Agent.

Use both areas together. A Source tells the system what to ingest; the Knowledge workspace lets you verify what was actually indexed.

43.2 Opening the Knowledge Workspace

From the Agent workspace, select Knowledge (what AI sees).

The page heading is What the AI knows.

The interface explains that every indexed page and Document is shown and that opening a row reveals the actual extracted text the model can use while answering visitors.

43.3 Knowledge Totals

The top of the page displays three primary totals:

  • Documents — indexed pages and files.
  • Chunks — embedding-sized blocks of text.
  • Total content — searchable text available to the AI.

These totals provide a quick indication of how much Knowledge has successfully entered the Agent's searchable index.

43.4 Searching Knowledge

Use Search by URL or title… to narrow the indexed Documents.

This is useful when the Agent has many pages or uploaded files and you need to verify one specific item.

43.5 Source Labels

Indexed Documents can identify their origin with labels such as:

  • URL
  • Sitemap
  • Notion
  • Google Doc
  • Pasted text
  • Auto-indexed visitor page
  • Uploaded file

HTML-based content may also show which crawler extracted it.

43.6 Opening a Document

Select a Document row to expand it.

The expanded view can display:

  • The Document title.
  • The indexed URL when available.
  • The number of Chunks.
  • The individual Chunk text.
  • Token and character counts.
  • An Open original link when the Document has a usable URL.

43.7 What a Chunk Represents

A Chunk is a smaller block of searchable content created from a Document. AIUNIFY Chat uses these blocks when retrieving relevant Knowledge for a visitor question.

Inspecting the Chunk text is one of the best ways to confirm whether the crawler or ingestion process captured meaningful content rather than navigation, empty markup, or unrelated text.

43.8 Indexing Didn't Finish

If a Document was crawled but has zero searchable Chunks, the Knowledge workspace can display Indexing didn't finish.

The interface explains that a transient embedding issue may be responsible and provides Reindex to retry.

43.9 Nothing Indexed Yet

If the Agent has no indexed Documents, the page displays Nothing indexed yet. and directs you to the Sources page.

43.10 Knowledge Review Best Practice

Do not evaluate Agent Knowledge only by the Source status. Open representative Documents and inspect the actual Chunks. The Knowledge workspace is the clearest user-facing view of what content the Agent has available for retrieval.

Write Your Comment