RAG
Knowledge base
A knowledge base is the corpus a retrieval system searches — the documents, chunks and embeddings that ground its answers.
Its quality bounds everything downstream. No retrieval strategy recovers information that was never indexed, and no prompt fixes a contradiction between two documents.
Freshness is the maintenance cost people underestimate. A knowledge base that drifts out of date produces confidently wrong answers that are harder to catch than obvious failures.
In practice
A retrieval index is a cache of your documentation, and every cache needs an invalidation story. The question worth answering before launch is how long a stale answer can persist after a document changes — a system confidently quoting last quarter's refund policy is worse than one that says it does not know.
Common questions
How often should I re-index a knowledge base?
As often as the source documents change, which usually means on write rather than on a schedule. The question worth answering before launch is how long a stale answer can persist after an update.
What makes a document hard to retrieve from?
Long unbroken sections with no headings, tables that lose meaning when split, and content that assumes context from elsewhere in the document. Structure that helps a human skim also helps a chunker.
Measure it
- Document Token ProfilerSection-by-section token share, with a density flag for dense content.
- Text Splitter for RAG and EmbeddingsToken-accurate chunking with configurable overlap. Respects sentence and paragraph boundaries.
- Embedding Cost CalculatorIndex build, re-indexing and query embedding, priced separately.