A Knowledge Base holds the content an AI Agent answers from: your documents, your websites, your FAQ lists. The content is split into chunks, embedded and indexed, and when a user asks a question the agent retrieves the most relevant chunks and grounds its answer in them.
Working with a Knowledge Base has three steps:
- Create it. Name the Knowledge Base and choose its embedding configuration. The embedding model cannot be changed after creation.
- Add data sources. Websites to crawl, text documents, visually rich PDFs, FAQ spreadsheets or numeric data tables. Sources are processed asynchronously, and each one shows its indexing status until it is ready.
- Connect it to an AI Agent with a Knowledge Retrieval tool. The tool selects the Knowledge Base and defines how many chunks each retrieval returns. A Knowledge Base must finish indexing before a tool can use it.
What a Knowledge Base is not. It is not a file store: uploaded documents are processed into searchable chunks, and the agent answers from those chunks rather than serving the original file back. It is also not a live mirror of its sources: a changed document must be uploaded again and indexed before answers reflect it.