How to
Knowledge
On this page
A knowledge collection is a set of files and web pages that an agent searches when it answers. Use one when the agent should answer from your own material, such as product documentation or an FAQ. You need the Manage agents permission.
Create a collection
- Open Admin console > Conclude Agents > Knowledge and select Add collection.
- Enter a Name and a Description, and select Create.
The drawer now shows Add websites and files. You can add both to the same collection.
Add files
Drop files on Drop documents here, or select Upload file. Conclude accepts Markdown, text, HTML and DOCX files of up to 10 MB each. It doesn’t accept PDF files.
A file is listed under its file name. To replace a file, delete it and upload the new version.
Add websites
- Enter an address, such as
docs.example.com/guide, and select Add website. - Check the Title. Conclude reads it from the page.
- To add every page below the address, tick Include every page under this URL. Ending the address with
*does the same. - Select the button at the bottom of the dialog. It says how many pages Conclude adds.
When you include every page, Conclude lists the pages from the site’s sitemap. It doesn’t follow links. Max pages and Max tokens limit how much it reads from the site.
Conclude follows the site’s robots.txt. A site that blocks crawlers can’t be added. A site without a sitemap can only be added one page at a time.
Keep sources up to date
Conclude indexes a page when you add it. It does not update the content automatically. After the content has changed, select Reindex on a page or Recrawl on a site. Recrawling preserves pages that have not changed.
The status below the collection’s name shows how many sources are ready, pending or failed. A failed source shows the reason below its name.
Add a collection to an agent
- Open the agent in Admin console > Conclude Agents > Agents and go to its Knowledge tab.
- Tick the collection and select Save.
On the Knowledge page, the Used by column shows which agents read each collection. An agent can read several collections, and several agents can read the same one.
How an agent uses a collection
The agent searches its collections on tickets and in channels. Each search returns up to five excerpts with their titles and links. The agent is told to cite the sources it uses, and to say so when the collection does not contain an answer.
An agent using an Anthropic model decides for itself when to search. For an agent using an OpenAI model, Conclude searches with the ticket or the latest messages and provides the excerpts to the agent before it answers.
Learning collections
Once a channel has a learning summary, the Knowledge page has two tabs, Collections and Learning. The Learning tab lists collections created automatically from channel tickets and conversations. You can’t add sources to these collections or assign them to agents. See Learning.
Limits
| Limit | Value |
|---|---|
| File size | 10 MB |
| Pages from one site | 200 by default, 1,000 at most |
| Tokens from one site | 1,000,000 by default, 5,000,000 at most |
All plans allow unlimited collections and sources. Indexing a source counts against the included quota. When the quota is used up, new sources remain available through keyword search, and it shows “Not embedded: the organization’s agent budget has been reached.”