Skip to main content
A Knowledge Source is structured, easily accessible information that Knowledge AI extracts from documents or external systems and provides to AI Agents. You can create Knowledge Sources from content types, such as PDF, CTXT, web pages, or via Knowledge Connectors that link to external knowledge bases. Knowledge AI breaks content into smaller units called Knowledge Chunks. You can also add Source metadata and Knowledge Source Tags to improve how AI Agents use the content.

Limitations

  • You can upload files with up to 10 MB as Knowledge Sources.
  • You can add up to 20 Source metadata pairs per Knowledge Source.
  • Source metadata pairs can have up to 1000 characters.

Knowledge Types

Knowledge AI organizes content using content types and Knowledge Connectors. While both provide information for your AI Agents, they are independent and use separate parsers.
Content types define the formats of content that can be used as Knowledge Sources. Supported formats include:Depending on the format of the file you use, you need to configure a specific parser to convert the content into Knowledge Sources.
Knowledge Connectors link your Knowledge Store to external systems, such as CRMs, help centers, or databases. They operate independently of content types and include their own parsers to fetch and transform data into usable knowledge for AI Agents. You can use prebuilt Knowledge Connectors from the Marketplace, such as Confluence and Diffbot, or create custom ones using the Extension framework.

Source Metadata

Source metadata contains information about the Knowledge Source and is formatted as JSON key-value pairs. The keys refer to the information the Source metadata holds and the value is the information, for example, publication: Elsevier. You can store Source metadata in the Input object or Context object and give your AI Agents access to the Source metadata through:
  • input.knowledgeSearch.topK[0].sourceMetadata in the Input object
  • context.knowledgeSearch.topK[0].sourceMetadata in the Context object
This way, AI Agents can provide relevant information about the source of their replies to users.

Examples

You can find examples of Source metadata in the following table:
KeyDescriptionMandatoryExample
titleThis key defines the documentโ€™s name. If a user requests the information source, the AI Agent can provide the documentโ€™s title in the response message.notitle: Cognigy xApps FAQs
urlThis key defines a URL as the documentโ€™s source. If a user requests the information source, the AI Agent can provide the link to the documentโ€™s source in the response message.nourl: https://www.cognigy.com
tagsThis key defines Knowledge Source Tags. To add more than Knowledge Source Tag, use an array. You can add up to 10 Knowledge Source Tags.no[cognigy, xapps, architecture]
You can add any Source metadata to a Knowledge Source and give AI Agents access to the Source metadata through input.knowledgeSearch.topK[0].sourceMetaData.{key}. For example, if you add the Source metadata author: Cognigy, where author is the key, and store it in the Input object, retrieve this information by using input.knowledgeSearch.topK[0].sourceMetaData.author.

Working with Knowledge Sources

  • GUI
  • API
  • CLI
You can view, create, edit, and delete Knowledge Sources in Build > Knowledge. Also, you can copy Knowledge Sourcesโ€™ Reference ID, disable and export Knowledge Sources, and download them as CTXT file.

Download Knowledge Sources as CTXT

You can download Knowledge Sources as a CTXT file, including Knowledge Sources created from web pages and other file formats. Downloading Knowledge Sources as CTXT files offers more flexibility in managing Knowledge Sources. With a CTXT file, you can:
  • Back up Knowledge Sources more easily after you have edited the Knowledge Chunks.
  • Use Knowledge Sources across Knowledge Stores with different embedding vector dimensions.
  • Edit downloaded Knowledge Sources more efficiently with external tools, such as text editors, with features like search and replace, bulk editing, and integration of AI tools.

More information

โŒ˜I