Introduction
Most organisations running AI cannot answer three questions: how many agents do we have, what are they doing with company data, and are they worth what they cost.
AI Governance answers all three, for every agent and model you run — the ones you built, the ones your vendors host, and the ones nobody registered.
"You cannot govern an agent you do not know exists. Everything here starts with finding out."
Where to find it
AI Governance is in the lower section of the platform side navigation. It opens the AI Governance overview: five groups of module cards, each with a live count — assets, agents, models, trace success %, open incidents, active policies, controls, running experiments, datasets, metrics, monthly spend and ROI.
From any module, click AI Governance in the breadcrumb to reopen the overview as a panel; Cmd/Ctrl-click opens it in a new tab.
Access is granted through platform roles. Ask your workspace administrator, or see Settings > Governance > Roles for how permissions are assigned.
Why a control layer
AI did not arrive through procurement. It arrived through a dozen platforms your teams already had, each with its own way of building agents and none of them talking to each other.
So the usual answer — govern each platform natively — produces a different policy model per vendor, no shared inventory, and no way to answer a regulator's question in one place.
AI Governance sits above the platforms rather than inside them. One inventory, one policy model, one incident trail, one cost and value picture, across every platform you have connected.
What it does
Find what you are running — Connect a platform and Discovery returns the agents and models it hosts — including the ones built by individuals that were never registered anywhere.
Bring them under governance — Register the assets that matter into the Agent and Model registries, with an owner and a workspace against each.
Apply your rules — Policies bundle controls and apply them to assets. Use the shipped framework packs, write your own, or both.
Investigate what happened — Every control match becomes an incident naming the asset, the policy, the control and the action — with the conversation behind it one click away.
Measure quality, cost and value — Score live traffic, test changes against fixed datasets, see where spend goes, and weigh it against the outcomes your agents produce.
How the module is organised
Group | Contains | Answers |
AI Assets | Discovery, Agent Registry, Model Registry | What do we have? |
AI Observability | Monitoring, Incidents | How is it behaving? |
AI Compliance | Policies, Controls | What rules apply? |
AI Evaluation | Experiments, Datasets, Metrics, Agent Optimizer | Is it any good? |
AI Cost & ROI | Cost Explorer, Outcomes & ROI | Is it worth it? |
What it is not
Worth being clear about, because the category is noisy.
Not an agent builder — It governs agents wherever they were built. Building them is elsewhere.
Not a replacement for platform-native controls — Where a hosting platform can stop something in the moment, that is the right place for a hard stop. This is the layer that spans all of them.
Not automatic — Discovery finds assets; you decide what to register, what to govern and what to enforce. Every gap between those is a deliberate step.
Where to start
New to the module — How AI Governance Works — the four stages end to end.
Want the object model — Core Concepts & Data Model.
Want to do something today — Govern Your First Agent.
Here for a specific job — Guided Paths, by role.