Glossary

Cache-augmented generation (CAG)

What is Cache-augmented generation (CAG)?

Cache-augmented generation (CAG) is an approach in which a language model does not search for knowledge with every question. Instead, a manageable knowledge base is loaded into its context window in advance and that state is cached, so answers need no separate retrieval step. The term was popularised by a late-2024 research paper as an alternative to retrieval-augmented generation (RAG). CAG suits small, stable knowledge bases; for large or frequently changing data, RAG usually remains more practical.

How KI.ordo helps

For knowledge, KI.ordo uses BM25 keyword search on the server and caches recurring prompt components with prompt caching to reduce costs.

Learn more →

Topics

RAG
Context window
Knowledge management

Try KI.ordo

Agents, flows and templates in one workspace – free 24-hour demo.

Start demo

← All terms

Related terms

Retrieval-augmented generation (RAG)

Retrieval-augmented generation combines search with text generation: before answering, the system searches a knowledge collection – such as manuals, policies or FAQs – for relevant passages and passes them to the language model as context. The answer is therefore based on current, company-specific content rather than only on the model's training knowledge. This reduces invented statements, makes sources traceable and allows updates without retraining.

RAG
knowledge search
+1

Large language model (LLM)

A large language model (LLM) is a neural network trained on very large amounts of text, enabling it to understand and generate language. It works by calculating the most likely next token, step by step. LLMs write, summarise, translate, answer questions and generate code; well-known examples include Claude, GPT, Gemini and Llama. Without access to current sources, they only know what was in their training data.

Language models
Generative AI
+1

Retrieval pipeline

A retrieval pipeline is the sequence of processing steps a system uses to find the right information for a question. Typical steps are: ingesting and cleaning documents, splitting them into chunks, building a search index, optionally rewriting the query, retrieving matches, ranking them (reranking) and passing the best ones to the language model. It is the foundation of every RAG application, and its quality largely determines how good the resulting answers are.

Information Retrieval

Information retrieval covers methods that find the documents or passages in a large body of text that match a query and rank them by relevance. Classic approaches such as BM25 weight search terms by how often and how rarely they occur; newer ones add semantic similarity. It underpins search engines, enterprise search and retrieval-augmented generation, in which a language model uses the retrieved passages as the basis for its answer.

Search
BM25
+1