VaultOS · The knowledge layer
Enterprise knowledge sits in a dozen systems, and none of your AI can act on it. VaultOS makes it one governed layer — plain Markdown under an open standard, catalogued in your own Google Cloud, grounding Gemini Enterprise with a citation on every answer, and written back to by people and agents alike. We run the discovery, the build and the handover.
In production · not a pitch deck
The stack
Three layers. Open formats at the base, your own cloud in the middle, grounded answers on top — and nothing proprietary holding it together.
Git-native, encrypted data layer
Markdown with YAML frontmatter under the Open Knowledge Format, in a Git repository you own. Every edit, update and deletion lands as a plain-text diff — auditable, reviewable, revertible. The sync server that keeps every device current only ever holds ciphertext.
Catalogued in your Google Cloud
The vault’s manifest lands in BigQuery. Knowledge Catalog registers every note automatically — governed, discoverable, and readable by agent tools. Scope isolation fails closed.
Cited answers, write-back, agents that check first
A dedicated search data store grounds Gemini Enterprise on your notes. Every answer cites the source. People and agents write decisions back, so the graph densifies instead of decaying.
The stall
Docs, decisions and runbooks live in SaaS tools, a couple of clouds, on-prem systems — and mostly in people’s heads. None of your AI can see across all of it.
Each initiative rebuilds the same missing context. Assistants answer from last week’s upload, not the organisation’s actual memory.
Without a governed layer, agents guess. There is no citation, no scope, and no way to ask what the organisation already decided.
Notes, decisions, and runbooks spread across tools and heads. Ours were too — hundreds of notes across six businesses.
Plain markdown in a linked graph — synced to every device, end-to-end encrypted, mirrored on a schedule.
The vault's manifest lands in BigQuery, and Knowledge Catalog registers it automatically — every note governed, discoverable, and readable by agent tools.
The full vault indexes into a search data store that Gemini Enterprise grounds on. Our last full sync: 1,038 documents, zero failures.
Ask in plain language — every answer cites the exact notes it came from. Scoped views for partners fail closed.
documents, grounded. Answers save back into the vault, our agents check policy before they act — and the morning brief cites the graph every day.
This is the exact pipeline we run internally. Enterprise builds get the same architecture on your own Google Cloud.
The vault, as a map
Notes, projects, dailies, indexes and agent memory — linked, not dumped in a drive. Same force layout, hit-testing and note drawer as the operating console. Click a node and the markdown file opens on the right. The live vault stays on a walkthrough.
Sample vault. Click a node to open its markdown.
Sample vault. Not Emerge’s operating graph. Click a node for the markdown. Reduced-motion devices get a static layout.
In production
Open files become a catalogued vault, then a permissioned index, then a cited answer. The sample graph above is the visual of that layer. The live operating graph is shown on a walkthrough. The stills on this page are labelled product illustrations.
Internal operating example: 1,038 documents indexed for Gemini Enterprise grounding on Emerge’s own environment, with zero import failures on the last full sync.
Knowledge tiles feeding retrieval, then a cited answer with a fail-closed permission gate.
One brain, two audiences
Humans ask in plain language. Agents check the same brain before they act.
Your team asks a question — “what did we decide about the refund policy in Q1, and why?” — and VaultOS answers from your own history, citing the exact notes. No hunting through drives. Every claim links to its source note, so answers are auditable — not guesses.
Agents retrieve only what their scope allows. A query outside that scope returns no sources rather than a best guess. Write-back lands as a reviewable diff, so the graph stays the record of authority.
Live demonstration
Public demonstration of cited answers on a governed vault. It is a demonstration, not your tenant.
Why an open layer
Running knowledge as governed files — rather than rows in someone else’s database — changes what compounds, and what you can walk away from.
Most AI knowledge tools require you to import your data into their database and read it back through their SDK. Here, the data never leaves plain Markdown in a repository you own — so the model, the editor and the vendor stay swappable. Including us.
Knowledge normally decays because writing it down is always someone’s side job. Here, people and agents both write decisions back into the vault as they work, so the graph gets denser every week instead of staler.
Siloed tools force the same facts to be re-entered in each one. Here, every person and every agent reads the same notes under the same permissions.
More agents normally means more decisions nobody is watching. Here, every note is catalogued and every change is a diff, so oversight scales with the fleet.
Leaving a knowledge platform is normally an export project with losses. Here it is one git clone — the files were always yours, and always readable without us.
Built for the enterprise
These are the controls running in production today — on our own vault. Enterprise installs use the same architecture in the customer’s Google Cloud project.
Enterprise builds deploy into your Google Cloud project. The vault, the catalog and the search data store stay inside your tenancy, under your billing and your controls.
Sync between devices is end-to-end encrypted, and the sync server only ever holds ciphertext — so the people operating it, us included, cannot read your notes.
Assistants sit behind access control with server-side token verification, and retrieval scopes fail closed: a query outside its scope returns no sources rather than a best guess.
Every change to the vault is a plain-text diff in Git, and every note is registered in Knowledge Catalog — so “what did we know, and when” is a query rather than an investigation.
Internal operating example
VaultOS is the production system behind Emerge’s own six-business portfolio — catalogued in Google Cloud, grounding Gemini Enterprise. Walk the sample graph on this page, ask the public demo, or book a walkthrough of the live vault.
1,038documents indexed for Gemini Enterprise grounding
0import failures on our last full sync
100%of answers grounded and cited to the exact note
6businesses on the same governed vault
Figures describe Emerge’s own environment, not a client result.
“We didn’t build VaultOS to sell it — we built it because we were drowning in our own knowledge. 1,038 documents across six businesses now answer our own questions, with citations, and our agents check the same brain before they act. Then clients started asking how.”Rami Alcheikh · Founder, Emerge Digital
Questions
A git-native convention: plain Markdown plus YAML front-matter. It is the canonical shape for knowledge AI agents can read. No SDK, no proprietary account — any LLM reads the files directly.
No. Obsidian is the default editor we configure because its graph view makes navigation intuitive, but your vault is plain markdown files. Any editor that opens text files — VS Code, iA Writer, Typora — works out of the box.
You do. Your vault lives on a server you control (a self-hosted CouchDB instance); the sync server only ever holds ciphertext — we cannot read your notes. If you opt into the grounded assistant, your notes are additionally indexed into a dedicated Google Cloud data store for your tenant — scoped, access-gated, and removable — and that choice is yours per tier.
Two Google Cloud layers, doing different jobs. The vault’s manifest lands in BigQuery and Knowledge Catalog registers it — that’s governance and discovery. Separately, the notes index into a search data store that Gemini Enterprise grounds on — that’s what makes answers cite the exact notes they came from. Answers can save back into the vault, closing the loop. It is the pipeline we run on our own six businesses.
Vault Studio (personal) is live within 3–5 working days of the walkthrough. VaultOS Team typically takes 1–2 weeks including the knowledge workshop. Enterprise OKF projects are scoped in a discovery engagement — timelines depend on the size of the knowledge base.
Yes. The monthly sync and agent subscriptions are cancel-anytime. When you cancel, your vault files stay with you — they’re on your server. Nothing is locked inside any Emerge system.
One git clone. The files were always yours. If your process needs a certified processor, the open format means your team can run the same architecture without us.
VaultOS grounds answers in the knowledge your team owns. The film uses an existing product illustration showing an example answer and its sources.
This silent film has no spoken dialogue.
On-screen text:
Get started
A free 30-minute walkthrough of a live VaultOS — the graph, the sync, the assistant that cites years of decisions. No deck, just the real system. Pick a time below.
Calendar not loading? Book on Cal.com or send an enquiry.