Start working with it. It keeps up.

What a turn holds, and what the company keeps.

Context is what one governed turn works with, and it is gone when the turn ends. Memory is what your company keeps: what happened, what we know, what we can do. Every turn deposits into memory, and memory folds back into the next turn. Practitioners are starting to call these two crafts context engineering and memory engineering; the diagram shows how ZeroH ships them as one governed loop. That loop is why nobody says "remember" in the sample week; the context travels on its own, across agents, with a receipt on every hand-off.

Drop a folder. Point at it. Done.

Standing up a governed RAG takes a folder you already have, one screen in Studio, and the agents you choose to attach it to.

1

Put the files in a folder

A SharePoint folder you already have, policies, contracts, standards, product docs. It is read where it lives; nothing gets converted or copied out.

→

2

Create the knowledge base in Studio

Point at the folder, say who can read it, and pick how its documents are cut. One screen, no integration project, no data pipeline to own.

New knowledge base

Who can read itEveryone can readA specific group

How its documents are cutGeneral documentsClause-numbered

Keeping it current

→

3

Attach it to any agent, or all of them

The help desk gets the product base, compliance gets the standards, everyone gets the handbook. Each agent then answers from what it is attached to, and from nothing else.

Help Deskproduct library✓

Compliancestandards✓

Operationsthe handbook✓

used by 3 agents · every answer cites this base

Governed RAG, out of the box.Cited answers, masked before any model, refused for anyone the folder does not let in, and it keeps getting better as your team uses it.

A real first ingest reads out like this: 34 files read, 1,208 chunks indexed, 2 skips named, each skip with its reason. sample figures

Already built a RAG? Keep it. Plenty of teams have already stood up an index, a graph in Neo4j or Graphiti, a vector store, a search cluster someone tuned for a year. Replacing it should not be the price of governance.

To be precise about what this is: your index becomes a source ZeroH can read under governance, while ZeroH keeps its own retrieval on the stack we measured. For a standards library, a graph store came out slower, pricier and harder to audit in that measurement.

A knowledge graph where every edge opens its proof.

The library compounds into a graph of what your company knows: the shelves, the documents, the decisions, and the connections the work itself created. Every edge is a stored fact or a deterministic join, and it opens the ledger row that proves it. The graph view is in build now; this rendition is illustrative.

The same build brings My Map, your personal view of the brain. Your raw files stay in your own OneDrive under Microsoft's ACL; ZeroH keeps a masked digest and a pointer, so your map is yours in the same way the company's memory is the company's.

The folder’s permission is the boundary.

Whoever can open the folder is whoever the answers reach. The folder permission is the only permissions model, so a document can only answer people who could have opened it.

Some shelves stay dark until you say otherwise. Salaries, invoices, board material: the assistant can know the shelf exists and still not be able to read the numbers on it, so it tells you, instead of guessing around them.