Skip to content

Knowledge bases

A knowledge base (KB) is a named collection of datasources — documents or web pages — that a flow can query for answers, either via a Simple Flow’s FAQ topic or a Graph Flow’s knowledge node / pre-hook knowledge_query integration lookup.

From a project’s Knowledge Bases page, create a KB and add datasources to it. A datasource is either:

  • A file upload — PDF, DOC/DOCX, Markdown, JSON, CSV, or XML.
  • A website URL — crawled and re-crawlable on demand.

Adding a datasource kicks off ingestion: the content is chunked, embedded, and written into Nairon’s vector store, with the datasource’s status moving through indexingactive (or flagged if it fails). Ingestion is asynchronous — the UI shows per-datasource progress, and you can re-index a datasource (e.g. after a source document changes) or the whole KB at once.

Retrieval at call time embeds the caller’s query and performs a hybrid similarity search over the KB’s chunks, returning the top matching passages for the agent to answer from.

  • Simple Flow: attach a KB to an FAQ topic; the agent answers from that KB’s content when the topic is matched.
  • Graph Flow: reference a KB from a knowledge-query pre-hook lookup, or from a dedicated knowledge node in the conversation graph, storing the retrieved passages into a variable for the agent to use.
  • Keep datasources focused and current — stale documents produce stale (or contradictory) answers.
  • Prefer several smaller, topic-scoped datasources over one enormous document; retrieval quality degrades with very large, unstructured single sources.
  • Re-index after editing a source document — Nairon does not detect external edits automatically for uploaded files.