Knowledge Step

Enrich with massive knowledge bases

This step allows for you to query against websites and documents for relevant data to inject into your prompts.

To use this, first add Knowledge sources in the Knowledge Tab, and then select all the sources you'd like to query against in the step.

By default, the most recent user message is what will be used to query the knowledge sources, but advanced users can change this to be any other data source available (i.e. you can pre-process the message with an LLM to achieve higher quality results.

Under the hood, this is a vector-lookup, so the most similar chunks to the searchy query will be output. You can adjust how many chunks, and how many different document can be used in the Advance Settings.

To make data available to the node, upload it in the Knowledge section (The book icon on the left navigation). Read more about Website Knowledge & Scraping, and Document Knowledge.