Use the Content Assistant
to ask questions about a single document in your repository. It provides accurate answers
within the context of the selected document content.
Before you begin
You need to set up the Content Assistant in your CPE and Content
Navigator environment. For more information, see the topic Setting up the Content Assistant.
About this task
After the Content Assistant
setup is complete, you can start asking questions to the assistant about a single document
in your repository.
Procedure
- Log in to IBM® Content Navigator by using your credentials.
Note: Use only HTTPS connection to access the IBM Content Navigator when the Content Assistant plugin is installed.
- Go to the object store where you want to use the Content Assistant.
- Select the document about which you plan to ask questions to the Content Assistant.
- Option 1 - Click Browse and select a single document in the object store.
You can ask questions about documents that are vector
indexed. The
icon
before a document name indicates that the document is vector indexed.
- Option 2 - Select an existing search or Click
in the
IBM Content Navigator console.
Execute the
search and then select a single document from the search results that is vector indexed. The
icon before a document name indicates that the document is vector indexed.
- Click the
icon on the Content Navigator web
console. The Content Assistant opens.
- Ask a question to the Content Assistant about the selected document
content.
By default, the Content Assistant
sends the entire document to the LLM to process your query and provide complete context. You can
change this setting to send only the most relevant document chunks instead by clicking the
icon in the Content Assistant and disabling the
Send entire document option. This approach is useful when the extracted text
exceeds 400,000 characters and the entire document cannot be passed as context.
For very large documents, sending only document chunks may be preferable. In this case, a vector
query then automatically retrieves the most relevant document chunks, which are then sent to the LLM
along with the prompt to generate a response.
Note: If your prompt is blocked and an error message is displayed, Granite Guardian input moderation has identified the content as potentially unsafe. Contact your administrator to review the moderation settings. For more information, see
Input moderation using Granite Guardian.
- Review the generated response.