---
schema: 1
type: term
title: Retrieval-augmented generation (RAG)
description: >-
  **Retrieval-augmented generation (RAG)** is a technique for giving a language
  model access to relevant external information at inference time, by retrieving
  data from an external source and providing it as context alongside the model's
  own knowledge. In this channel's material, RAG appears as the typical
  implementation of an agent's **memory** component — the part of an agent that
  supplies connected data and context to the reasoning engine
  [^50a3b3fa-9ad2-59a3-9c76-ddfe74635066@220].
canonical: >-
  https://liquidgold.cc/channel/everyinc/explore/term/retrieval-augmented-generation-rag-aac94dd9f247
markdown: >-
  https://liquidgold.cc/channel/everyinc/explore/term/retrieval-augmented-generation-rag-aac94dd9f247.md
llms: null
parent: 'https://liquidgold.cc/channel/everyinc/llms.txt'
channel: everyinc
updated: '2026-10-23T22:12:35.346Z'
language: null
children: []
---


# Retrieval-augmented generation (RAG)

Canonical: https://liquidgold.cc/channel/everyinc/explore/term/retrieval-augmented-generation-rag-aac94dd9f247

Parent: [Channel index](https://liquidgold.cc/channel/everyinc/llms.txt)

## Definition

**Retrieval-augmented generation (RAG)** is a technique for giving a language model access to relevant external information at inference time, by retrieving data from an external source and providing it as context alongside the model's own knowledge. In this channel's material, RAG appears as the typical implementation of an agent's **memory** component — the part of an agent that supplies connected data and context to the reasoning engine [^50a3b3fa-9ad2-59a3-9c76-ddfe74635066@220].

## Where RAG appears in this channel **Retrieval-augmented generation (RAG)** shows up in Silvio Savarese's four-part decomposition of an agent. Breaking an agent into its components, he identifies **memory** — typically implemented with RAG — alongside a brain (the reasoning engine), actuators (API and function calls), and an interface.  [^50a3b3fa-9ad2-59a3-9c76-ddfe74635066@220] ## Why the memory part matters Savarese's broader argument in the same interview is that **an agent without connected data and context is useless**. Enterprise deployments stumble on **siloed data** — sales and marketing manage data differently, and AI cannot operate well when the dots aren't connected.   [^50a3b3fa-9ad2-59a3-9c76-ddfe74635066@342] [^50a3b3fa-9ad2-59a3-9c76-ddfe74635066@485] In that framing, RAG is the technique serving the agent's *memory* slot: it is how an agent retrieves relevant information from its connected data so the reasoning engine has context to work with.  - RAG is the typical implementation of an agent's **memory** component in Savarese's four-part decomposition. - Its role connects directly to the channel's central enterprise-AI challenge: agents need connected data and context to be useful. - Siloed enterprise data is the practical obstacle that memory/retrieval systems like RAG exist to work around.