Wikis and help centers are knowledge bases for people. An AI knowledge base has different requirements: clean text, meaningful chunks, retrieval that ranks by relevance, and an interface a model can call. Bolting RAG onto a human wiki usually disappoints — the content was never prepared for machines.
What makes a knowledge base AI-ready
- Every document normalized to Markdown, whatever it started as — PDF, DOCX, HTML, a scan, a URL.
- Hybrid retrieval: semantic search catches meaning, keyword search catches exact terms; the blend beats either alone.
- Citations back to the source document, so answers are checkable.
- Machine access: REST API for pipelines, MCP for agents like Claude.
How it works in Kit for AI
Create a knowledge base, ingest files and URLs into it, and it's immediately searchable — and chattable, with cited answers generated from your own documents. Your agents query it over MCP with one config block; your code hits the same API.
Start small
Ingest ten documents your team actually references and ask real questions against them. A knowledge base earns trust by answering correctly with citations — scale comes after.