AI Governance Dashboard
- Tier: Ultimate
- Offering: GitLab.com
- Status: Limited availability
The availability of some tiles and cards on this page is controlled by feature flags. For more information, see the history.
Security and compliance teams can use AI Governance Dashboard to monitor AI agent activity across a group.
This feature is in beta. It is subject to change without notice. For more information, see GitLab Testing Agreement.
AI Governance Dashboard surfaces key performance indicators and data cards that give you visibility on:
- How AI agents are being used.
- Which developers are most active.
- Which projects have the most exposure.
The dashboard shows data for GitLab Duo Agent Platform (DAP) agents and, when the Show control offers it, for connected agents. Connected agents are external coding agents such as Claude Code, Cursor, and OpenCode that report their sessions to GitLab. By default, the dashboard covers the last 7 days. You can change the date range to the last 30 days.
Prerequisites:
- You have the Owner role for the top-level group, or a custom role with the
read_agent_artifactsability.
View the dashboard
- In the top bar, select Search or go to and find your top-level group.
- In the left sidebar, select AI > Governance.
- Select the Dashboard tab.
Filter the dashboard
Two controls at the top of the dashboard filter the tiles and most cards:
-
Show: The agents to include:
- All agents (default): GitLab Duo Agent Platform agents and connected agents.
- DAP: GitLab Duo Agent Platform agents.
- Connected: Connected agents.
-
Date range: Last 7 days (default) or Last 30 days.
Each card that ignores the date range says so below its list.
Key performance indicator (KPI) tiles
The dashboard header shows up to three KPI tiles. Each tile displays a count for the selected date range and a sparkline showing the daily trend. The tile also shows the change compared to the previous period of the same length.
Total number of agents
The Total number of agents tile shows the number of distinct active agent instances in the selected date range. An agent instance is unique per user, project, namespace, agent type, and environment. A connected agent instance is unique per user, project, agent type, and agent identity. GitLab Duo Agentic Chat conversations are excluded from this count.
Total number of agent sessions
The Total number of agent sessions tile shows the total number of agent workflow sessions in the selected date range. This includes all DAP workflow types: IDE, web, chat, and ambient sessions. Agentic Chat is included. With All agents or Connected, the tile also counts sessions of connected agents.
Total number of tool calls
The Total number of tool calls tile shows the number of times agents invoked a tool in the selected date range. Every invocation counts, so a tool called 10 times adds 10. The tile counts tool calls from GitLab Duo agents and from connected agents that report them, such as Claude Code.
Data cards
Below the KPI tiles, data cards provide breakdowns of agent activity.
Which AI agents developers are using
The Which AI agents developers are using card shows a ring chart of the distinct developers who used AI agents in the selected date range. The total is in the center of the chart. The slices depend on the Show control:
- With All agents, one slice per agent product, such as GitLab Duo, Claude Code, Cursor, or OpenCode.
- With DAP, one slice per GitLab Duo Agent Platform flow type, such as chat, code review, or software development.
- With Connected, one slice per connected agent type.
Hover over a slice to see its developer count and its share of all developers. A developer who used two agents is counted in both slices, so the shares can add up to more than 100%.
Select View all agents to open the group’s Agents page.
Top tools used by agents
The Top tools used by agents card lists the 10 tools that agents invoked most often in the selected date range, with the number of calls for each. The card follows the Show control and the date range.
Select View tool management to open the Tool management tab, where you set what each tool is allowed to do.
Tool calls by risk
The Tool calls by risk card shows a ring chart of the tool calls in the selected date range, grouped by the risk level of the tool, as set in Tool management:
- Read
- Write
- Destructive
- Unclassified, for tools that the tool registry does not know.
The total number of calls is in the center of the chart. The legend shows each level’s share.
Select View tool registry to open the Tool management tab.
Most recent sessions
The Most recent sessions card lists the latest agent sessions, newest first. It ignores the date range. It follows the Show control.
The card lists sessions from GitLab Duo and from connected agents. Each row shows the agent that ran the session, such as GitLab Duo, Claude Code (glab), Cursor (glab), or Claude Code (Compliance API), and the number of audit events the session produced.
Select a GitLab Duo session to open it in the AI audit event report. Select a connected agent session to open the report without opening a session. Select View all audit logs to open the full report.
Most used agents
The Most used agents card lists the agents used most in your group over the last 30 days, with the project each agent belongs to. The ranking always covers 30 days, whatever date range you select.
Most active developers using agents
The Most active developers using agents card shows the top users by agent session count over the selected date range. Use this to understand which developers are most actively using AI agents.
Projects with the most active agent sessions
The Projects with the most active agent sessions card shows the top projects by agent session count over the selected date range. Use this to identify which projects have the highest volume of AI agent activity.
MCP servers
The MCP servers card lists the Model Context Protocol (MCP) servers registered for your group, with their status. Use this to see which external MCP servers your agents can reach.
To read a server’s full description, hover over or focus the truncated description text.
The card lists registered servers only. It does not show how often each server is used.