The governance frame

Agents need the decision rights people never wrote down.

Small companies run on shared understanding. Somewhere around the mid-market, who decides what is true was never written down. People could work around that. Agents can't. Once agents do real work, the decision rights have to be explicit.
Four questions

Every governed AI program answers the same four questions.

Answer them once, in the layer every agent reads, and you don't have to answer them again for every new tool.
01 · Who decides what's true?

Approval before use

Each area of company knowledge has a named owner. With stewardship on, changes wait for the owner's approval before any agent may rely on them. Agents propose, people approve.

Without it: Whatever was edited last, or whatever the agent wrote itself, becomes the answer.
02 · Who may see what?

Scoping by role

Each agent gets a key scoped to the nest its job needs, and restricted documents reach only the people named on them. A support agent's nest doesn't hold finance, and one client's nest never holds another client's data, whatever the prompt says.

Without it: “Don't touch client data” is a request in a prompt. Agents apologize after they break it.
03 · What changed, and when?

Versioning

Every change is kept with who made it and who approved it. When a policy changes, agents get the new approved version, and the old one stays in the history.

Without it: Last year's policy sits next to this year's, and both match the question.
04 · Who read what, and what did it say?

A record of every read

Every read is recorded: which person or API key, which document, and when. The version history is hash-chained, so you can rebuild what the approved knowledge said on any past date.

Without it: When legal or a customer asks, nobody can say what the AI was working from.
Look closely

Three things that sound like governance, and aren't enough on their own.

An admin on/off switch

Turning an assistant's memory off for the company is a control. It doesn't say which knowledge an agent may treat as true.

Guardrails on actions

Agent platforms approve what an agent may do. Very few govern what it reads before it decides.

File permissions

Agents inherit the user's access. An over-shared folder is over-shared to every agent that user runs.

Build it or buy it: agents on governed context →

Decision rights, made explicit

Owners approve their own area. Nobody becomes the bottleneck.

Knowledge areaOwner approvesAgents that may read it
Product and pricing factsProduct leadSales, marketing and support agents
Policies and processOperationsAll internal agents
Customer accountsAccount ownerThat customer's agents only
Finance, legal and HRFunction headExplicitly allowed agents only

An illustrative layout. Your areas and owners will follow how your company already works.

See the controls behind the frame.

Role-based access, audit trails, conversation monitoring and model governance, in detail.

Governance controlsTrust and security

Map your decision rights with us.

Bring one knowledge area and the person who owns it. A short call sets up approval and scoping on documents like yours.

Get a free consultationStart a 30-day trial