Four years of IP. One queryable intelligence layer. This is the most critical build of the sprint — the moat that makes everything else irreplaceable.
Most people who use AI tools are querying the entire internet. Every answer they get is contaminated by everything the model has ever seen — billions of web pages, other people's ideas, other people's voices, other people's wrong answers. The output is generic by design. It has to be, because it knows nothing specific about you.
The Vault changes that completely. Instead of querying the internet, every AI agent in this ecosystem queries a private, sovereign database built exclusively from four years of Cyber Coastlines IP. Methodology frameworks. Narrative architecture. Brand voice documentation. Course materials. Business SOPs. Every meaningful document ever produced under this signature — ingested, indexed, and made instantly retrievable.
When Claude reasons inside this system, it doesn't free-range. It reads your documents, retrieves the most relevant sections, and reasons exclusively against that context. The output sounds like you because it's built from you. That's not a feature. That's the entire strategic advantage.
The vault is the moat. Nobody can replicate what lives in there because nobody has four years of this specific IP. Every other tool in the stack is replaceable. The vault is not.
Obsidian is where the raw IP lives. Every document, every framework, every piece of content that has ever defined this ecosystem gets organized here in plain markdown files — local on the MacBook, backed up with end-to-end encryption via Obsidian Sync. Nobody at Obsidian can read the content. It never passes through a third-party cloud unencrypted.
Before the sprint begins, the vault content needs to be organized into six buckets: Creative Operations Methodology, Brand & Voice, Causeway Nova narrative assets, The Creative Companion curriculum, Business Operations, and Session Intelligence. This organization is the pre-work. The consultant cannot build the RAG on a pile of unstructured files.
Obsidian is the source of truth. Everything else in the vault architecture is downstream of it.
LlamaIndex is the librarian. It reads every document in the Obsidian vault, breaks them into semantically meaningful chunks, converts those chunks into vector embeddings, and stores them in Supabase. When a query comes in — from Claude, from a Lindy agent, from a direct question — LlamaIndex retrieves the most relevant chunks and assembles them into context before the reasoning engine ever sees them.
This is what makes the system intelligent rather than just searchable. A keyword search finds documents that contain the word. LlamaIndex finds documents that contain the meaning — even when the exact words don't match. That distinction is the difference between a search engine and an intelligence layer.
LlamaIndex is open source and free. The consultant should configure the ingestion pipeline to run automatically whenever new content is added to the Obsidian vault — ensuring the intelligence layer stays current without manual intervention.
Supabase hosts the intelligence layer — the embedded, indexed, queryable version of the vault. Every chunk that LlamaIndex processes gets stored here as a vector embedding alongside its source metadata. When retrieval happens, Supabase performs the vector similarity search that returns the most contextually relevant results in milliseconds.
The hosting decision for Supabase — managed cloud at $25/month or self-hosted on a private VPS — should have been made before the sprint began. If it hasn't been, that decision happens before a single line of vault architecture is written. Both options work. The self-hosted path gives full sovereignty. The managed path gives faster setup. The consultant's recommendation drives this call.
Either way, the vault data belongs to Cyber Coastlines LLC. The embeddings, the indexes, the configuration — all of it lives in GitHub, all of it is reproducible, all of it is owned.
The vault is not a single tool — it's a pipeline. Understanding the flow is as important as understanding the individual components. Here is what happens from the moment content is created to the moment intelligence is returned.
The Obsidian vault needs to be organized before the ingestion pipeline can run. These are the six top-level folders. Every piece of existing IP belongs in one of them. If something doesn't fit, it either needs a new bucket — which requires a conversation during the sprint — or it's not ready for the vault yet.
Module 02 is not complete until the vault returns relevant results on real queries. Not test queries with obvious answers — real questions about the CreativeOps methodology, the Causeway Nova narrative, the Creative Companion curriculum. If the retrieval isn't working accurately, the Reasoning Engine module that follows will be built on a broken foundation. Fix it here. Everything downstream depends on this being right.
Module 03 connects the reasoning engine. Claude meets the vault — and everything it produces from this point forward is grounded in four years of your IP.