Your AI already knows the world.
An NKB holds what it doesn’t.

A Neural Knowledge Base is a folder of plain Markdown files, a map of how those files relate, and a place an AI companion works inside with you. The files are the source of truth. The map tells the AI what else bears on what it is reading. The place gives it identity and continuity from one session to the next.

The knowledge your AI does not already have. Models learn public material in training: public law, public science, public code. They never saw your clients, your methods, your decisions or your history.

An NKB holds that private knowledge in a form an AI can stand in, so every session starts informed instead of starting over. That is where it earns its keep.

Because it proves scale. NKB Law maps every US statutory section, about 1.23 million of them, with millions of relationship edges. We built it to show the substrate holds at that size, on public data anyone can check.

It is an exhibit, not the product. Your NKB uses the same machinery for knowledge that only you have.

We measured it. Curated notes plus an AI that searches them beat embedding-based retrieval of the same private corpus, on every engine we tested. The results are published: Measuring Our Own Claim.

We also published what did not help, and what is still untested.

On the surface they look alike: both are folders of linked Markdown notes. But they are built for different jobs. An Obsidian vault is an app for a person to read and write notes. An NKB is built for an AI to read, reason and work inside the notes.

What makes the difference sits above the files: identity, protocols, memory between sessions, and retrieval that was measured against RAG. A vault could hold the same files and still have none of that.

Yes. NKB notes are plain Markdown with YAML frontmatter, which Obsidian reads. It is a fine way to read and edit your notes by hand.

Two things to know. NKB connections are stored as ids in each note’s frontmatter, not as [[wiki links]], so Obsidian’s graph view will not show most of them. And Obsidian users report that very large vaults, past about 100,000 notes, index slowly and the graph view stops working. A personal NKB is far smaller than that.

Yes. OKF asks for Markdown files with YAML frontmatter and one required field, type. We audited our own NKB on 2026-08-08: all 1,580,690 notes conform. Everything an NKB adds sits on top of that open floor, so your files stay portable.

Not automatically. Obsidian reads Markdown with YAML properties, so it can read OKF files. But it does not require frontmatter or a type field, so a typical vault is not an OKF bundle as it stands.

Community plugins close the gap, for example one that checks a vault against the spec and one that exports to OKF. Obsidian itself makes no OKF claim that we could find.

A starter NKB: a small set of files with room to grow, plus a boot prompt. Paste the prompt into ChatGPT, Claude, Gemini or Grok, or download the files and work in Claude Code, Cursor or Grok CLI, where the AI can read and write the map directly.

The Seed is free, with no credit card. The only ask: point others to registration rather than passing the Seed around.

Your durable identity in the IntentGate ecosystem. You receive one when you register for a Seed, and it is the key to what is built on the substrate as the Federation grows.

You do. A Seed is plain files you download and keep. You can read every one, change them, and put them under version control. Nothing about your NKB is locked in a vendor’s memory.