Skip to main content
What your agent knows beyond its instructions is set in the agent editor, in the Background knowledge section: uploaded PDFs, imported websites and web search. The editor saves everything automatically. It takes effect from the next conversation.

Documents as knowledge (PDF)

Under Documents (PDF) you upload PDFs whose text the agent uses as background knowledge. The full text of all files is appended to the end of the system prompt. The agent answers factual questions on this basis and treats the content as its own knowledge. It does not mention that it has documents and does not name file names.
1

Name the agent

Uploading only works once the agent exists. For a new agent, give it a name first.
2

Upload PDFs

In the Background knowledge section, drop the PDFs onto the field or click it. You can select several files at once.
Uploaded PDFs and imported websites share the list Agent knowledge. Show prompt text shows the text the agent gets from an entry.
If your agent runs on an OpenAI Realtime model, upload and import are locked. These models do not include documents in the system prompt.
When you duplicate an agent, the uploaded PDFs are copied too.

Import website as knowledge

Under Import website as knowledge, the editor reads the text of a website and appends it to the system prompt as full text, like a PDF.
1

Enter the address

Enter the address of the website, for example example.com. You can leave out https://.
2

Set the scope

Maximum subpages sets how many pages the import loads at most (up to 10). In the focus field you name terms such as references, projects: pages whose address contains one of them are fetched first.
3

Import

Click Import. An import can take a few minutes. After that the website appears in the list Agent knowledge.
The import only reads text that is in the page’s HTML. Pages that load their content with JavaScript or reject the request yield no text.
With web search, the agent can look up current information on the internet during a conversation — for example opening hours, news, or prices that are not in its background knowledge. Turn it on in the Background knowledge section under Web search with the Web search checkbox. It is off by default. Web search works with all language models except OpenAI Realtime. With the default model gemma-4-31b-it and with Cerebras, GPT-5.4 mini runs the search in the background; a search then takes a few seconds before the agent answers.

Transfer an agent

You can hand an agent over to another Avaluma account, for example after setting it up for a client.
1

Open the menu

In the Agents overview, open the agent’s menu (three dots) and choose Transfer….
2

Enter the recipient

Enter the email address the recipient uses to sign in to Avaluma and click Transfer.
After that, the agent belongs to the recipient and disappears from your list. Agent ID, configuration and uploaded PDFs stay the same, so links, embeds and kiosk URLs keep working. Only the new owner can give the agent back, by transferring it to you. The transfer is rejected if
  • there is no Avaluma account with this email address,
  • you enter your own address, or
  • the recipient has already reached the agent limit of their plan (see Pricing & Billing).