Dev.to
7/14/2026

A Folder of Docs Is Not a Knowledge Base
Short summary
A folder of markdown docs fails as a knowledge base because it retrieves prose chunks, not atomic facts, and loses provenance once embedded. The fix is a pipeline that distills sources into individual sourced claims before any agent queries them. The author's agentproto corpus-cli tool implements this with import, distill, validate, and serve steps that keep every claim traceable to its origin.
- •Docs-as-knowledge-base fails because chunks aren't facts and provenance is lost after embedding
- •Solution: distill sources into atomic, sourced claims before retrieval
- •agentproto corpus-cli provides a four-command pipeline to build, validate, and serve a queryable corpus
Generated with AI, which can make mistakes.
Is this a good recommendation for you?



