Adding and editing knowledge base entries

What the knowledge base is for

When the flow's AI Assistant step answers a customer, it searches the knowledge base. Every entry you add is an answer the agent can give. On a topic the knowledge base doesn't cover, the agent doesn't make an answer up.

Open it from “Knowledge base” in the menu. All members can see entries; adding, editing and deleting require the owner or admin role.

Adding a new entry

Press “New entry” and choose what to add: “Text”, “Web page”, “File” or “Image”.

For text:

1. Write the title: the topic, or the question a visitor would ask (“How long does delivery take?”).
2. Write the content. It supports bold, italic, lists and links. Both the title and the content are required.
3. Choose the “Scope”: “Shared (all agents)” or one specific agent.
4. Save. The agent uses the entry from the next question on.

Adding a web page

Enter the address of a page on your website (FAQ, pricing, contact); the agent reads the text on it. The entry is named after the page's own title, which you can change later.

• Under “Updates”, choose how often the page is read again: every day, every week, every month or never. The agent learns what changes on the page.
• If the page fails to open several times in a row, the agent stops reading it and the entry shows a “Page not opening” badge in the list. It comes back on its own once the page opens again.
• “Update now” in the row's three-dot menu reads the page again right away.
• Only public pages can be read; pages that need a login or load their text later may have no text to read.

Adding files and images

Choose “File” or “Image” and drag the files in, or pick them with “Choose files”. You can pick several; each becomes its own entry, named after the file.

• File: PDF, Word, Excel, slides, text and HTML files. The agent reads the text inside. Scanned PDFs (where the text is a picture) have no text to read; upload a version with text.
• Image: JPG, PNG, WebP and GIF. The agent reads the text in the image and what it shows; good for a menu, price list or poster.
• Up to 10 MB per file.
• Files and images are never shown to visitors; they are only a source the agent reads when answering. Open the entry to see the text the agent read.

Finding, editing and deleting entries

Use the filters above the list to search by title, and to filter by entry type, scope and who can see it.

Clicking a row opens the entry read-only. To edit or delete, use the three-dot menu on the row. The agent stops answering from a deleted entry.

Tips

• You don't need to split long documents: the agent finds the part of the document that relates to the question and uses it. Short facts still work best as separate entries.
• See the questions the agent couldn't answer in the “Unanswered questions” card of agent analytics, and fill the gaps from there.
• The agent searches your entries by meaning, not only by words (this approach is called RAG). It usually finds the right entry even when the question uses different words or another language; for prices, names and codes it also looks for the exact word.