Skip to content

Knowledge Base

What it's for

The Knowledge Base holds the business facts your AI agents can look up during calls. Add your hours, services, prices, policies, and FAQs as entries, or upload documents. Then sync them to AI memory. Your agents search this memory during calls so they can give callers accurate answers. Each workspace has its own knowledge. It is never mixed with other workspaces.

The page has four tabs: Entries, Knowledge Graph, Search, and Ingest Documents.

Add a knowledge entry

  1. Open the Entries tab.
  2. Click Add Entry.
  3. Choose a Type: Business Info, Product, Service, FAQ, or Policy.
  4. Enter a Title, such as "Business Hours".
  5. Enter the Content. Write it the way you want the agent to understand it.
  6. Optionally, add Tags (comma-separated), such as hours, schedule.
  7. Click Add Entry.

New entries show a yellow Pending badge until you sync them.

Sync entries to AI memory

Entries only reach your agents after you sync them.

  1. On the Entries tab, click Sync to Memory (N). N is the number of pending entries.
  2. Wait for the confirmation message. Synced entries show a green Synced badge and the sync date.

To push every entry again, click Resync All. Use this if you think memory is out of date.

The stat cards show Total Entries, Synced to Memory, and Pending Sync.

Edit or delete an entry

  1. Find the entry. Use Search knowledge base... or the type menu (All Types, Business Info, and so on) to narrow the list.
  2. To edit, click the pencil icon, make your changes, and click Save Changes.
  3. To delete, click the trash icon.

Warning

Editing an entry sets it back to Pending. Click Sync to Memory again so your agents get the update. Deleting an entry happens right away, with no confirmation.

Upload documents

  1. Open the Ingest Documents tab.
  2. Under Documents, choose one or more files. You can use PDF, DOCX, TXT, or MD files.
  3. Optionally, adjust Chunk size (chars) and Chunk overlap (chars). The defaults are 1000 and 100. Smaller chunks (about 500 to 800) with 10 to 15% overlap give more precise answers. Larger chunks keep more context together.
  4. Optionally, turn on LLM context (better retrieval) and pick an LLM model. This adds a short summary to each chunk so facts link together better. It adds a small cost per chunk.
  5. Click Ingest N file(s).

An Ingestion Summary table lists each file, how many chunks it produced, and its status (ok, partial, skipped, or an error).

Note

Uploaded documents are processed in the background. They can take a moment to appear in the Knowledge Graph and Search tabs.

Test what your agent will find

  1. Open the Search tab.
  2. Type a question a caller might ask, such as "what are the business hours?"
  3. Click Search or press Enter.

Results are ranked facts, the same way your agent sees them during a call. Each one has a relevance bar. If nothing comes back, you see No matching facts. Add or sync more knowledge, then try again.

Explore the knowledge graph

The Knowledge Graph tab draws the people, things, and facts the AI pulled from your knowledge and past calls.

  1. Click Load Graph. Click Refresh later to reload it.
  2. Drag nodes to rearrange them. Use the zoom controls in the bottom-right corner.
  3. Use Search graph… at the top right to highlight matching nodes, or pick a type from the menu next to it.
  4. Open Entity Types at the top left to show only certain types: Entity, Document, Memory, Conversation, or Tag. Click Clear all to reset.
  5. Click a node to see its toolbar:
    • Inspect opens details, its Citations / Facts, and Related nodes.
    • Neighbors focuses on the nodes connected to it.
    • Hide removes it from view. Click Show all to bring hidden nodes back.

If you see No graph data yet, sync some entries or upload documents, then refresh.