Knowledge Base / RAG

SaaS, open source and skills for knowledge base / rag work — with the limits of each written down.

Before you pick

What decides it
Your corpus is mostly near-duplicates — drafts, revisions, superseded policies. Similarity alone can't rank them; ask what breaks the tie: recency, ownership, version.
What to avoid
They respect permissions but end obscurity. Sharing links set too broadly years ago become one question away, and the cleanup lands after go-live.
What is changing
Retrieval is becoming a planned loop: the model splits one question into several searches, reranks each, then merges. Slower and costlier per answer.

Open source

Repositories you run yourself.
Search an area, or a tool by name.