GitLab Duo Agent Platform Impact dashboard

  • Tier: Premium, Ultimate
  • Offering: GitLab.com, GitLab Self-Managed, GitLab Dedicated
  • Status: Beta

This feature is in beta. Share your feedback in issue 632127.

The GitLab Duo Agent Platform Impact dashboard shows how your organization uses Agent Platform and what that usage costs. The dashboard displays aggregated data from your groups and projects into metrics for user adoption and engagement, work activity, and credit consumption.

Use this dashboard to:

  • Measure the volume of Agent Platform workflows across your organization.
  • Track credit consumption by model, flow type, and group or project.
  • Monitor user adoption trends, including retention, churn, and activity tiers.
  • Identify which projects and groups are most actively using Agent Platform.

Overview

The Overview tab displays a high-level summary of Agent Platform usage for the selected scope and time period.

  • Active projects: Number of projects with at least one Agent Platform session.
  • Active users: Number of users with at least one Agent Platform session.
  • Completed sessions: Number of created Agent Platform sessions that completed.
  • Credits consumed: Number of credits consumed by Agent Platform sessions.
  • Where credits went: Number of credits consumed by each Agent Platform flow type.

Adoption

The Adoption tab displays metrics on how users are adopting Agent Platform features for the selected scope and time period.

The Users section displays the following metrics:

  • Total users: Number of users with at least one Agent Platform session.
  • New vs returning: Number and percentage of new and returning users with at least one Agent Platform session. Returning users are users who had an Agent Platform session in the previous period of the same length.
  • Users by activity: Number of users by Agent Platform flow type. A user of multiple flow types counts in each category.

The Engagement section displays the following metrics:

  • Multi-activity users: Number of users with sessions for multiple Agent Platform flow types.
  • Abandonment: Number of users who were active on only one day in the previous period and did not return in the selected period.
  • Sessions by tier and top capabilities: Number of sessions by user tier for the five most used Agent Platform capabilities.

The Adoption tiers section displays the following metrics:

  • Users across tiers: Number of users with at least one Agent Platform session.
  • Total sessions: Number of Agent Platform sessions across flow types and agents.
  • Share of users vs share of sessions: Number of users and sessions by user tier.
  • Session intensity by tier: Number of users and sessions by user tier.
  • Group and project comparison: Number of users and used credits by group and project.

Work

The Work tab displays metrics on how Agent Platform features are being used for the selected scope and time period.

The Activity summary section displays the following metrics:

  • Sessions: Total number of Agent Platform sessions.
  • Sessions by activity: Number of Agent Platform sessions by activity.
  • Sessions over time: Number of Agent Platform sessions created, stacked by flow type.

The MR cycle time section displays the following metrics:

  • Cycle time: Duo-created: Median and 75th percentile time in days from open to merged for merge requests created by GitLab Duo. Includes the number of merged merge requests.
  • Cycle time: all other: Median and 75th percentile time in days from open to merged for merge requests that were not created by GitLab Duo and have human authors. Includes the number of merged merge requests.

A merge request is created by GitLab Duo when an Agent Platform session records that it opened the merge request, whoever the author is. A merge request that a session was started from, for example to review it, counts as not created by GitLab Duo.

The all other metrics exclude merge requests by bots, service accounts, and other non-human users, so they compare GitLab Duo with work done by people.

The Merge request volume section displays the following metrics:

  • Opened by Duo: Number of merge requests created by GitLab Duo.
  • Merged: Duo-created: Number of merged merge requests created by GitLab Duo.
  • Merged: all other: Number of merged merge requests that were not created by GitLab Duo and have human authors.
  • Duo created merge requests by status: Number of merge requests created by GitLab Duo with current status over the selected period.

The Capability outcomes section displays the following metrics:

  • Agentic Code Review sessions: Number of sessions of the Code Review flow.

  • Reviews requested: Number of reviews requested to Code Review.

  • Reviews published: Number of reviews published by Code Review.

  • With comments: Number of reviews published by Code Review with at least one comment.

  • Without comments: Number of reviews published by Code Review with no comment.

  • Agentic Code Review outcomes over time: Number of reviews requested to and published by Code Review with and without comments over the selected period.

  • Agentic Developer sessions: Number of sessions of the Developer flow.

  • Completed: Number of completed sessions of the Developer flow.

  • Paused: Number of paused sessions of the Developer flow.

  • Canceled: Number of canceled sessions of the Developer flow.

  • Failed: Number of failed sessions of the Developer flow.

  • Agentic Developer sessions over time: Number of completed, paused, canceled, and failed sessions of the Developer flow over the selected period.

  • Failed by reason: Number of failed sessions of the Developer flow, grouped by the reason recorded when the session failed:

    • no_activity: The session was stopped by the stuck-session cleanup.
    • no_status_reported: The CI workload ended without reporting a final status.
    • job_failed: A CI job did not succeed.
    • workload_not_created: The CI workload for the session could not be created.
    • preconditions_not_met: No eligible runner was available for the session. Before GitLab 19.6, this reason was also recorded when the namespace had no compute minutes left.
    • compute_minutes_used_up: No runner was available because the namespace has no compute minutes left.
    • could_not_prepare_session: GitLab could not prepare the session.
    • start_failed: The session could not start.
    • api_reported: The agent reported an error.
    • token_rejected: The GitLab Duo Workflow Service rejected the session’s token and did not run the session.
    • tool_rules_unavailable: The GitLab Duo Workflow Service could not load the tool rules for the session and did not run the session.
    • Unknown: No reason was recorded. This mostly affects sessions that failed before reasons were recorded.
  • Agentic Fix Pipeline sessions: Number of sessions of the Fix Pipeline flow.

  • Suggestions posted: Number of fix suggestions posted by GitLab Duo on failed pipelines.

  • Suggestions applied: Number of applied fix suggestions posted by GitLab Duo on failed pipelines.

  • Agentic Fix Pipeline suggestions over time: Number of fix suggestions posted by GitLab Duo and applied on failed pipelines over the selected period.

  • Agentic SAST sessions: Number of sessions of the Agentic SAST flow.

  • MRs created after finding vulnerability: Number of merge requests created by Resolve SAST Vulnerability sessions.

  • Agentic SAST sessions over time: Number of SAST flow sessions created and merge requests created in the sessions.

Spend

The Spend tab displays metrics on how much Agent Platform features are being used for the selected scope and time period.

  • Credits by model, by capability: Number of credits used by different models, stacked by Agent Platform capability.
  • Credits: Total number of credits consumed.
  • Chat: Number of credits consumed by Chat sessions.
  • Credits over time: Number of credits used over time, stacked by capability.

On GitLab Self-Managed, credit data syncs every six hours and requires a paid license. The first sync imports up to 90 days of credit data. On larger instances, the first sync can take several runs to complete.

View the dashboard

Prerequisites:

  • The Owner role for the group to view credit usage data.
  • For GitLab Self-Managed, ClickHouse must be configured and Siphon replication must be enabled.
  1. In the top bar, select Search or go to > Explore.

  2. In the left sidebar, select GitLab Intelligence.

  3. Select DAP Impact.

  4. Select the tabs to view the different metrics.

  5. Optional. By default the dashboard displays metrics for the top-level group in the last 30 days. You can filter the results:

    • To view metrics for specific groups or projects, in the respective tab, from the Scope dropdown list, select the groups or projects you want to view.
    • To view metrics for a different time period, in the respective tab, from the Period dropdown list, select the period you want to view.
    • To reset the selected scope and period, select Reset.
  6. To analyze the displayed metrics with GitLab Duo, select Analyze with Duo.

    GitLab Duo Chat opens with the Data Analyst Agent, which reads the dashboard data. The agent must be turned on for your default GitLab Duo namespace. If the agent is turned off, Chat can’t analyze the dashboard and displays a warning.