Know every AI agent in your tenant.
Copilot Studio made it easy for anyone to publish an agent. Module B produces the list nobody has — every agent, its owner, its status and its risk flags — and keeps it current, read-only.
Nobody decided to build an ungoverned AI estate
It happened one useful agent at a time. The first thing a governance board needs is not a policy — it is a list.
Orphaned agents
An agent whose owner has left, or whose owner no longer resolves in the directory. It keeps answering, and nobody is accountable for what it says.
Unused agents
Published, reachable, and not used in months. Each one is an unreviewed surface still exposed to the people it was shared with.
Sensitive-data agents
Agents grounded in knowledge sources that touch financial, legal or HR content — the ones whose answers matter most and get reviewed least.
Shadow agents
Built by a maker, never registered with anyone, never put through a review. You cannot govern a list nobody has ever produced.
Ownership concentration
When most of the estate belongs to one or two people, their departure is a governance incident with a date on it.
Drift since last scan
What appeared, what disappeared and what changed hands since the last snapshot — the only honest measure of whether sprawl is getting better.
Inventory, decide, track
Three screens: the list, the queue of agents needing a decision, and the trend across scans.
Every Copilot Studio agent in the tenant
A live, read-only inventory of every agent — owners, channels, status and knowledge sources — built automatically rather than maintained by hand in a spreadsheet nobody updates.
- Published, draft and unused agents separated
- Owner and channel per agent
- Knowledge sources listed, not guessed at
The agents that need a decision
Each flagged agent becomes a line in the same recommendation list as the licence actions — review this one, retire that one — so governance work lands in a queue rather than a report.
- Orphaned and unused agents surfaced as actions
- Priority per row
- Export to CSV or PDF for the governance board
A baseline, then a trend
Scan again whenever you need a fresh picture and compare snapshots. Agent sprawl is only visible as a direction of travel — one count tells you nothing.
- Snapshot history per tenant
- Agent count and risk flags over time
- Evidence a governance board will accept
It inventories agents. It never touches them.
Copilot Insights cannot alter, disable or delete an agent — the consent has no write scope. Every flag is a recommendation for the person who owns the governance decision.
No changes to an agent
Not its publishing state, not its knowledge sources, not its sharing. The product observes; your administrators act.
No conversation content
Agent metadata only. What people asked an agent, and what it answered, is never read or stored.
Revoke in one click
Remove consent from Microsoft Entra whenever you like, and delete the organisation and its scan data with it.
Inventory first, policy second
Copilot Insights · Module B
The standing inventory: which agents exist, who owns them, which are dormant or orphaned, and which touch sensitive content. Refreshed on demand, read-only, available the day you sign in.
- Automatic agent discovery across the tenant
- Risk flags per agent
- Recommendations and exports
- Trend across snapshots
AI governance & readiness service
What you do with the list: an approval model for new agents, a lifecycle for retiring old ones, sensitivity labels and DLP behind the knowledge sources, and the executive roadmap around it.
- Agent lifecycle and approval model
- Responsible AI and executive roadmap
- Sensitivity labels, DLP and retention
- Oversharing remediation before grounding
Produce the list, then govern it.
One free, read-only scan gives you every Copilot Studio agent in the tenant, with the orphaned, unused and sensitive ones already flagged.
Read-only · agent metadata only · never conversation content
More on Copilot Insights
Every page of the Copilot Insights sub-site, and the way back to the rest of LogiSam.