Knowledge base

The knowledge base is where you store the information your AI uses to answer customers. You add articles (manually, by scraping a website, or by uploading files), organize them into groups, and then attach those groups to channels. When the AI drafts a reply, it searches the attached groups for relevant content and grounds its answer in what it finds. This guide walks through every step, from adding your first article to wiring it up so the AI actually uses it.

You'll find the knowledge base at Settings → Knowledge Base (route /knowledge-base). Managing the knowledge base requires the OWNER or ADMIN role; agents and viewers can generally read it.

How the knowledge base fits in#

Think of the knowledge base as the AI's reference library. On its own, an article doesn't do anything. It becomes useful through a three-step chain:

1. Build articles — add the content you want the AI to know.
2. Group them — bundle related articles into groups (for example, "Shipping", "Returns", "Product FAQ").
3. Attach groups to channels — in Settings → Channels → channel → AI & Knowledge, you attach one or more groups to a channel.

Once a group is attached to a channel, every time the AI drafts a reply on that channel it searches the attached groups for content relevant to the customer's message and uses what it finds to ground its answer. This technique is called RAG (retrieval-augmented generation): the AI retrieves your real content first, then writes its answer based on it, instead of relying only on what the model already knows.

This is the key distinction worth remembering: the knowledge base is what grounds AI replies. (FAQs, by contrast, are primarily for the website chat widget's greeting screen — see FAQs.)

Adding articles#

You can add articles three ways. Each produces an article with a source badge so you can tell at a glance where it came from.

1. Write manually

Choose to write an article by hand when you want full control over the wording.

• Provide a title and the content.
• The article gets a MANUAL source badge.
• Manual articles are the only type you can later edit directly (see Editing and deleting).

Manual articles are ideal for policies, canned explanations, and anything that doesn't already live on a web page or in a document.

2. Scrape a website URL

Point the knowledge base at a URL and it will crawl the site and turn the pages into articles.

• The crawler follows links and ingests up to roughly 50 pages.
• Resulting articles get a URL source badge.
• Great for importing an existing help center, documentation site, or marketing pages in one shot.

Tip: scrape a focused section of your site (for example, your help center subdomain or a docs path) rather than your entire marketing site, so the AI's reference material stays relevant.

3. Upload a file

Upload a document and the knowledge base extracts its text into an article.

• Supported formats: PDF, TXT, MD, DOCX.
• Maximum file size: 10 MB.
• Resulting articles get a FILE source badge.
• Useful for product manuals, internal guides, and policy documents you already maintain as files.

Processing statuses#

Scraped and uploaded articles aren't ready instantly — the content has to be fetched and processed first. Each article shows a status:

• PROCESSING — the content is being fetched and prepared. The article isn't available to the AI yet.
• READY — processing finished successfully. The article is now searchable and can ground AI replies. You can view its content.
• ERROR — something went wrong. The article shows an error message explaining what happened (for example, a page that couldn't be reached or a file that couldn't be parsed).

If an article gets stuck in ERROR, check the message, fix the underlying cause (a bad URL, an unsupported or corrupt file), and re-add it.

Organizing with groups#

Groups are how you bundle related articles so you can attach them to channels as a unit.

• Articles can be organized into groups.
• You can filter the article list by group to focus on one topic at a time.
• You can delete a group. When you do, its articles aren't deleted — they simply become ungrouped.

Because you attach groups (not individual articles) to channels, your grouping is also your targeting strategy. For example, a sales channel might get a "Pricing" group while a support channel gets "Troubleshooting" and "Returns".

Views and finding articles#

The article list offers two layouts:

• Grid view — cards, good for scanning visually.
• List view — compact rows, good for working through many articles.

Combine the views with the group filter to navigate large knowledge bases quickly.

Batch operations#

When you have many articles to manage at once, use batch selection:

• Select multiple articles, or use select-all / clear to manage the whole filtered set.
• Move to group — reassign the selected articles to a different group in one action.
• Delete — remove the selected articles.

Batch operations are the fastest way to reorganize after a large scrape or upload.

Editing and deleting#

• Edit — you can edit manual articles (their title and content). Scraped and uploaded articles reflect their source, so re-import them to update their content.
• View — open any READY article to read its full content.
• Delete — remove an individual article, or several at once via batch operations.

Stats dashboard#

The knowledge base includes a stats dashboard so you can see the state of your library at a glance:

• Total article count.
• Ready count (articles available to the AI).
• Processing count (still being prepared).
• A group breakdown showing how articles are distributed across groups.

Use the ready vs. processing counts to confirm an import has finished before you rely on it in replies.

Recommended workflow#

Here's a reliable order to set up a knowledge base from scratch:

1. Gather your content. Decide what the AI needs to know — policies, product details, common how-tos.
2. Add it. Scrape your existing help center or upload your manuals to get bulk content in fast, and write manual articles for anything that isn't already documented.
3. Wait for READY. Check the stats dashboard and confirm articles have moved from PROCESSING to READY. Fix any ERROR articles.
4. Group it. Create groups by topic and assign articles (use batch "move to group" to do this quickly).
5. Attach groups to channels. Go to Settings → Channels → channel → AI & Knowledge and attach the relevant groups to each channel. See Channels.
6. Test a reply. Send a test message on the channel and confirm the AI's draft reflects your articles. If it doesn't, re-check that the right group is attached and that the articles are READY.

Tips for good articles#

• One topic per article. Smaller, focused articles retrieve more accurately than long catch-all pages.
• Use clear, descriptive titles. Titles help both you and the retrieval step find the right content.
• Write the way customers ask. Include the words and phrasings customers actually use, so their questions match your content.
• Keep it current. Outdated articles produce outdated answers. Edit manual articles when policies change, and re-scrape or re-upload when source content changes.
• Group by how you'll target it. Since groups are what you attach to channels, organize them around the audiences and channels you'll serve.

Related#

• Channels — attach knowledge-base groups to a channel in the AI & Knowledge tab.
• AI replies — how the AI uses your knowledge base to draft replies.
• AI providers — configure the model that powers replies.
• FAQs — the widget greeting chips, and how they differ from the knowledge base.
• Getting started — set up your organization and channels first.
• Glossary — definitions of RAG, groups, channels, and more.