Knowledge base

A knowledge base is the organized collection of records your assistant can search and reference, such as meeting notes, decisions, contacts, and sources, kept together so an answer can come from your own material.

A knowledge base is the organized collection of records your assistant can search and reference, kept together so an answer can come from your own material instead of the model's general training. It holds the durable things your work depends on: meeting notes, decisions, contacts, project files, and saved sources, each one labeled and findable. Say you ask your assistant what you agreed about a refund policy. With a knowledge base in place, it searches your decision log, finds the entry, and answers from the record you wrote, citing where it came from. Without one, it guesses from generic patterns.

Builder example

Telling an assistant to answer from your own records only works if those records sit in one searchable place. A support bot pointed at a tidy knowledge base of past tickets and policy notes can resolve a password reset question by quoting your actual steps. Point the same bot at scattered files in five apps and it either misses the answer or invents one. Decide what belongs in the knowledge base, keep each record labeled, and the quality of every answer rises with it.

Common confusion: A knowledge base is the stored collection of records, while retrieval is the act of searching that collection to pull the relevant ones into an answer. You build and curate the knowledge base once; retrieval runs every time you ask a question.