Skip to main content
The Knowledge Base is where you give your agent its expertise. Everything the agent knows about your business comes from the sources you add here. The more relevant and well-organized your sources, the better your agent answers. Open it from the agent sidebar under Expertise → Knowledge Base. Its subtitle sums up the goal: “Create organized content collections that help your AI provide consistent, detailed responses.” To add content, select New Knowledge Item — this opens a tabbed screen with five source types.

​
Source types

You can train an agent on five kinds of sources. Mix and match as many as you need.

Text

Paste plain text directly — notes, snippets, policies, anything you can type.

Files

Upload documents like PDF, DOCX, and TXT — manuals, policies, guides, reports.

YouTube

Pull the transcript from a YouTube video and train on what’s said.

Q&A

Write a question and its exact answer — best for FAQs and canned responses.

Links (Websites)

Add website links and index their content automatically.
The Knowledge Base screen with the New Knowledge Item action, usage meters, and source list.

​
Usage

The Knowledge Base header shows how much of your plan’s capacity you’ve used:
MeterWhat it tracks
Characters UsedTotal characters indexed across all sources (for example, 0 / 15.0M).
Documents UsedNumber of sources added (for example, 0 / 100).
These limits depend on your plan — the values above are an example from a free/trial plan. Check the meters on your own Knowledge Base for your current allowance.
You can filter the source list by Search by name, KB items, Status, Sort by, and Date Range to find a specific source quickly.

​
How processing works

Adding a source kicks off background processing that turns your content into something the agent can search:
1

Ingest

The file or URL is read and the raw text is extracted (including from PDFs and Word documents).
2

Chunk

The text is split into small, overlapping passages so the agent retrieves only the relevant part of a long document.
3

Embed & index

Each passage is converted into an embedding and stored in your agent’s private vector index, ready to be searched on every question.
You don’t have to wait on this screen — processing runs in the background, even for large files. See Manage sources to track status and retrain.

​
Source statuses

StatusMeaning
Unprocessed / PendingQueued, waiting to start.
ProcessingBeing read, chunked, and indexed.
CompletedIndexed and live — the agent can now use it.
FailedSomething went wrong. Reprocess it or remove and re-add it.
Only Completed sources are used in answers. If your agent ignores content you just added, check that the source finished processing.

​
What makes a good knowledge base

  • Prefer clean text. Searchable PDFs and well-structured docs beat scanned images and screenshots.
  • One topic per source where you can — it keeps retrieval sharp.
  • Add Q&A pairs for the questions you already know customers ask. They give precise, predictable answers.
  • Remove stale content. Outdated sources produce outdated answers.

​
Searching your knowledge base

You can search across an agent’s indexed content directly from the Knowledge Base screen to confirm a topic is covered before you rely on it in a live agent.

Upload a file

Manage & reprocess sources

How answers are generated

Test your agent