Source: Notebook experiment 14: RAG + CAG
Runnable scaffold that preloads stable company facts (CAG) and retrieves volatile per-order facts (RAG).
Contents
- What: runnable experiment in the owner’s public Notebook repo
- Author: Loic (coolloic)
- Date: undated (read 2026-11-14)
- Original: experiments/14-rag-cag
Summary
Combines cache-augmented generation for a small, always-needed core of facts with retrieval for a larger volatile corpus, using prompt caching in production.
Key claims
- CAG removes a retrieval round-trip for stable, always-relevant context.
- Cached context still costs tokens on every call and can go stale, so cache only what is stable and bounded.