What does RAG actually do?
RAG stands for retrieval-augmented generation. The idea is straightforward. A system searches a corpus of text for passages relevant to your question, places those passages into the model's context window, and the model generates an answer grounded in what it found. Without RAG, the model answers only from training data, which has a cutoff date and no access to your personal reading.
RAG is a technique, not a product. It describes a pattern: retrieve, then generate. The quality of a RAG answer depends entirely on what is in the corpus and how retrieval selects passages. Chunk the corpus poorly and the model gets irrelevant context. Chunk it well and answers improve dramatically. The technique is only as good as its source material.
What makes a knowledge base personal?
A personal knowledge base is a curated library of sources you chose to save. Unlike a generic corpus assembled by a crawler, every item reflects something you read, highlighted, or noted. KnowMan reads the full text of what you save, not just the title. Answers cite the saved sources so you can check them. Memories keep decisions with their reasons, so the context behind a choice is never lost.
Free plan includes unlimited saving, keyword search, highlights, notes, Memories, and MCP connection. Import from Pocket, Instapaper, Omnivore, Readwise, Wallabag, or browser bookmarks. Pro adds semantic search, AI summaries, rules that sort articles, newsletter subscriptions, and export to Obsidian, Logseq, or EPUB.
When does a curated library beat generic chunking?
Generic chunking splits text into fixed-size pieces and indexes them for retrieval. It works well for broad searches across large, impersonal datasets. But chunk boundaries can also split an argument in half, and a generic corpus holds no record of what you decided. When you need sources you trust, decisions kept with reasons, and a library that serves both you and your AI agents, a curated knowledge base wins.
- Sources you selected and trust, not a random web crawl
- Decisions stored with their reasoning, searchable later via kb_memories
- The same library serves you in the app and your agents via MCP
- Full-text reading, not just titles or snippets, so context is preserved
How do AI agents read your knowledge base?
KnowMan exposes an MCP server with six tools: kb_search (search everything saved), kb_get_item (read full text of a saved item), kb_ask (ask a question, get an answer with citations), kb_memories (read decisions, playbooks, facts, notes), kb_recent (see recent saves), and kb_save (save a new link).
Works with Claude Desktop, Claude Code, Codex, Cursor, and any MCP client. Setup takes two steps: create an agent key in Settings with only the permissions you choose, then paste one config snippet into the client. Every new account starts with 14 days of Pro, no credit card required.
The honest framing: RAG describes how answers are built. A personal knowledge base describes what those answers are built from. You need both—the retrieval technique and the curated source material. KnowMan gives you the source material, structured so any MCP-compatible agent can retrieve from it.
Questions people ask
Is RAG the same as a knowledge base?
No. RAG is a retrieval technique—pull relevant passages, feed them to a model, generate an answer. A knowledge base is the curated collection those passages come from. You can apply RAG over a knowledge base, but they solve different problems.
Can I use RAG without a personal knowledge base?
Yes. RAG works over any text corpus. But answers will only be as good as that corpus. A personal knowledge base gives you sources you selected, decisions stored with reasons, and highlights you made—things a generic corpus cannot provide.
How do AI agents access my KnowMan library?
Through the MCP server. Create an agent key in Settings with chosen permissions, paste the config snippet into Claude Desktop, Claude Code, Codex, Cursor, or any MCP client. Six tools let agents search, read items, ask questions, check memories, see recent saves, and save new links.
Does KnowMan use semantic search?
Pro plan includes semantic search—search by meaning, not just keywords. Free plan offers keyword search over saved text, highlights, and notes. Both plans include Memories and MCP connection.