Knowledge bases

A knowledge base gathers the content your agents search to answer customers: terms and conditions, customer service FAQs, product sheets, internal procedures, help center pages. Each piece of content is split into segments and indexed, so an agent finds the passage that answers a customer's question in milliseconds, instead of guessing the answer.

This page explains how to create a base, feed it and keep it up to date. To have an agent use it, see Connect a base to an agent.

Before you beginCopy Anchor Link

• Bases are managed from Knowledge in the main menu (page title: Knowledge base).
• To import from Mayday, Zendesk, Stonly or Mindsay, the matching integration must first be connected in Settings > Connectors.

Create a baseCopy Anchor Link

1. Open Knowledge.
2. In the left column, click New.
3. Enter the base name (for example "Customer service FAQ"), then confirm with Enter or the confirm button.

The new base opens right away. Each base has its own address in the app (/knowledge/<id>): you can share it with a colleague or find it again after a reload.

Move between basesCopy Anchor Link

• The left column lists your bases, each with the number of documents it contains.
• The Filter bases… field finds a base by name, ignoring case and accents ("general" finds "Général").
• Bases are shown 15 per page.
• The ↑ and ↓ keys move from one base to the next when the cursor is not in an input field.

Add contentCopy Anchor Link

Click Add under the base name. The Add to base panel offers eight source types.

TextCopy Anchor Link

1. Choose Text.
2. Fill in Document title and the content in Content to index….
3. Optional: enter a Source URL (optional), shown with the document.
4. Click Index.

The title and content are required.

FileCopy Anchor Link

1. Choose File.
2. Click Choose or drop a file, or drag and drop the file onto the area.

The content is extracted (slide text, sheets and cells…) then indexed automatically. The file name becomes the document title, and a badge shows its format in the list.

The format is checked before upload. A file in another format, without an extension or empty is rejected with an explicit message, and the panel stays open so you can choose another one. A file that is too large is rejected with the message "File too large to be imported."

Question / answer CSVCopy Anchor Link

CSV suits FAQs: each row becomes a document.

1. Choose CSV.
2. If needed, click Download the CSV template to start from a ready-to-fill file.
3. Click Choose a CSV file (or drop the file).

The file must be comma-separated and have three columns:

Indexing of a CSV continues after upload: the new documents appear in the list after a few seconds. If they don't, click the refresh icon next to Documents.

WebsiteCopy Anchor Link

Flowtative explores the site from the address you give, discovers the linked pages, then indexes them.

1. Choose Website.
2. Enter the start address (for example https://example.com/help).
3. Optional: a Title (optional) to recognize the source.
4. Choose the Refresh mode (see below).
5. Adjust Max pages (100 by default, 0 = no limit) and Depth (2 by default, 0 = start page only, 10 at most).
6. Click Crawl.

Each site you add appears under Web sources with its status, the number of pages, the frequency and the date of the last sync.

While an exploration or indexing is running, no other action is possible on the source.

Configure a website sourceCopy Anchor Link

Hover over the source, then click the Configure icon to open the Configure the source panel.

The Advanced section adds: Allowed external domains (when external links are followed), Allowed internal subdomains, HTML tags to exclude (e.g. nav, footer, to skip menus and footers), Country (proxy) and Credit budget (0 = ∞).

The same panel offers three one-off actions, available depending on the source status:

• Re-explore now: reads the site again to find new and modified pages (once the source is up to date).
• Index new pages: only indexes pages not yet indexed.
• Re-index everything: re-indexes all pages with the current settings, for example after changing cleaning or language.

Click Save to apply the settings.

Mayday, Zendesk, Stonly and MindsayCopy Anchor Link

1. Choose the tool.
2. If no integration is connected, the panel points you to Settings > Connectors. Connect your workspace, then come back.
3. Select a collection in Choose a collection… (for Mindsay: Choose a Mindsay base…). The number of documents in each collection is shown.
4. Click Import the collection (Mindsay: Import the base).

Attached collections appear under External collections. For each one, you can:

• click the card to show only its documents (a collection · … filter appears above the list; click it to remove it);
• Synchronize to fetch the latest changes;
• Remove from base (confirmation Remove?).

For Mindsay, a real-time integration is shown by the real time label instead of the sync date.

Track your documentsCopy Anchor Link

The Documents section lists the content of the base, 10 per page. Each document shows its title, an excerpt, the number of segments and status badges.

Click a document (or View content) to open its detail:

• the source URL, if any;
• the Processing errors, with their date and message;
• the Content, in Rendered mode (formatted reading) or RAW (exact indexed text);
• the Indexed segments, that is the passages the search can return.

To delete a document, click the Delete document icon.

Test the searchCopy Anchor Link

Before handing the base to an agent, check that it returns the right passages.

1. In the Test semantic search… field, type a question the way a customer would ask it, for example "How long do I have to return an item?".
2. Click Search.

Up to 5 results are shown, each with the document title, an excerpt and a relevance score between 0 and 1. Only passages scoring at least 0.3 are kept. If nothing comes up, the message "No segment found." is shown: rephrase the question, or add to the base.

Base propertiesCopy Anchor Link

Click Properties at the top right of the base.

Three actions sit at the bottom of the panel:

• Re-index: recomputes indexing for all documents.
• Empty the base: permanently deletes all documents. The base, its web sources and its integrations remain.
• Delete the base: permanently deletes the whole base, with its documents, web sources and support tool attachments.

To empty or delete a base, you must type its exact name before confirming. Both operations are irreversible.

Best practicesCopy Anchor Link

• One base per scope. For example, keep "Customer service FAQ" and "Product catalog" apart: a support agent doesn't need technical sheets, and more targeted results give better answers.
• Short, titled documents. One question per document in a FAQ, an explicit title: the search returns more precise passages.
• Test with your customers' words. Reuse real phrasings from your transcripts rather than internal vocabulary.
• Limit site exploration. Use Included paths and HTML tags to exclude to keep only useful content (help center, product pages) and skip menus, blog and legal notices.
• Choose automatic refresh for sites that change often, to avoid outdated answers.

Learn moreCopy Anchor Link