Retrieve GitLab Duo and SDLC trend data

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

Use the GraphQL API to retrieve and export GitLab Duo data.

Retrieve AI usage data

  • Offering: GitLab.com, GitLab Self-Managed, GitLab Dedicated

The AiUsageData endpoint provides raw event data. It exposes Code Suggestions-specific events through codeSuggestionEvents and all raw event data through all.

On earlier versions with GitLab Duo Pro, the AiUsageData endpoint returns null without an error message.

You can use this endpoint to import events into a BI tool or write scripts that aggregate the data, acceptance rates, and per-user metrics for all GitLab Duo events.

Data is retained for three months for customers without ClickHouse installed. For customers with ClickHouse configured, there is currently no data retention policy.

The all and codeSuggestionEvents attributes have a maximum date range of one month. If you need data spanning multiple months, run separate queries for each month.

The all attribute is filterable by startDate, endDate, events, userIds, and standard pagination values.

To see which events are being tracked, you can examine the events declared in the ai_tracking.rb file.

The GitLab Duo Chat request_duo_chat_response event does not populate the extras field. Unlike Code Suggestions events, Chat interactions do not carry language or suggestion metadata. An empty extras object on this event is expected behavior. For the events that carry Chat token counts, see GitLab Duo Non-Agentic Chat token usage.

For projects and groups

For example, to retrieve usage data for all Code Suggestions events for the gitlab-org group:

graphql
query {
  group(fullPath: "gitlab-org") {
    aiUsageData {
      codeSuggestionEvents(startDate: "2025-09-26") {
        nodes {
          event
          timestamp
          language
          suggestionSize
          user {
            username
          }
        }
      }
    }
  }
}

The query returns the following output:

graphql
{
  "data": {
    "group": {
      "aiUsageData": {
        "codeSuggestionEvents": {
          "nodes": [
            {
              "event": "CODE_SUGGESTION_SHOWN_IN_IDE",
              "timestamp": "2025-09-26T18:17:25Z",
              "language": "python",
              "suggestionSize": 2,
              "user": {
                "username": "jasbourne"
              }
            },
            {
              "event": "CODE_SUGGESTION_REJECTED_IN_IDE",
              "timestamp": "2025-09-26T18:13:45Z",
              "language": "python",
              "suggestionSize": 2,
              "user": {
                "username": "jasbourne"
              }
            },
            {
              "event": "CODE_SUGGESTION_ACCEPTED_IN_IDE",
              "timestamp": "2025-09-26T18:13:44Z",
              "language": "python",
              "suggestionSize": 2,
              "user": {
                "username": "jasbourne"
              }
            }
          ]
        }
      }
    }
  }
}

Alternatively, to retrieve usage data for all GitLab Duo events for the gitlab-org group:

graphql
query {
  group(fullPath: "gitlab-org") {
    aiUsageData {
      all(startDate: "2025-09-26") {
        nodes {
          event
          timestamp
          user {
            username
          }
        }
      }
    }
  }
}

The query returns the following output:

graphql
{
  "data": {
    "group": {
      "aiUsageData": {
        "all": {
          "nodes": [
            {
              "event": "FIND_NO_ISSUES_DUO_CODE_REVIEW_AFTER_REVIEW",
              "timestamp": "2025-09-26T18:17:25Z",
              "user": {
                "username": "jasbourne"
              }
            },
            {
              "event": "REQUEST_REVIEW_DUO_CODE_REVIEW_ON_MR_BY_AUTHOR",
              "timestamp": "2025-09-26T18:13:45Z",
              "user": {
                "username": "jasbourne"
              }
            },
            {
              "event": "AGENT_PLATFORM_SESSION_STARTED",
              "timestamp": "2025-09-26T18:13:44Z",
              "user": {
                "username": "jasbourne"
              }
            }
          ]
        }
      }
    }
  }
}

For instances

  • Offering: GitLab Self-Managed, GitLab Dedicated

Prerequisites:

  • You must be an administrator for the instance.

For example, to retrieve all GitLab Duo usage events for the entire instance:

graphql
query {
  aiUsageData {
    all(startDate: "2025-09-26", endDate: "2025-09-30") {
      nodes {
        event
        timestamp
        user {
          username
        }
        extras
      }
    }
  }
}

The query returns the following output:

json
{
  "data": {
    "aiUsageData": {
      "all": {
        "nodes": [
          {
            "event": "CODE_SUGGESTION_SHOWN_IN_IDE",
            "timestamp": "2025-09-26T18:17:25Z",
            "user": {
              "username": "jasbourne"
            },
            "extras": {}
          },
          {
            "event": "AGENT_PLATFORM_SESSION_STARTED",
            "timestamp": "2025-09-26T18:13:44Z",
            "user": {
              "username": "johndoe"
            },
            "extras": {
              "session_id": "abc123"
            }
          }
        ]
      }
    }
  }
}

GitLab Duo Non-Agentic Chat token usage

  • Offering: GitLab Self-Managed, GitLab Dedicated

GitLab Duo Non-Agentic Chat records the tokens it uses as AI usage events. The token counts are in the extras field, which the aiUsageData query returns only for instances.

Chat records these events:

  • REPORT_DUO_CHAT_TOKEN_USAGE: Recorded for each agent turn that the AI Gateway reports token usage for. A question can take several agent turns. The extras field contains the request_id and a usage object that maps each model to its input_tokens and output_tokens.
  • COMPLETE_DUO_CHAT_RESPONSE: Recorded once for each response. The extras field contains the request_id and, if the response failed, its error_code.

The agent turns and the response for one question share a request_id.

Only agent turns are recorded. When a Chat tool sends its own request to a model, for example to read an issue, those tokens are not recorded. Slash commands do not use agent turns and record neither event. When GitLab Duo answers a merge request comment that mentions @GitLabDuo, the answer uses agent turns and records both events. Requests for a code review record neither event, and neither do answers from the Code Review Flow.

The token counts are the values the AI Gateway reports. If a self-hosted model does not report token usage, the AI Gateway estimates the counts. The REPORT_DUO_CHAT_TOKEN_USAGE event does not indicate that the counts are estimates.

Turn off token usage recording

By default, GitLab records token usage for GitLab Duo Non-Agentic Chat. When you turn it off, GitLab stops recording REPORT_DUO_CHAT_TOKEN_USAGE events. GitLab still records COMPLETE_DUO_CHAT_RESPONSE events and keeps existing events. If AI logs are on, they still record token counts.

Prerequisites:

  • Administrator access.

To turn off token usage recording:

  1. In the upper-right corner, select Admin.
  2. In the left sidebar, select GitLab Duo.
  3. Select Change configuration.
  4. In the Data and privacy section, under Record GitLab Duo Non-Agentic Chat token usage, clear the checkbox.
  5. Select Save changes.

You can also use the duo_token_usage_tracking_enabled application setting.

Retrieve AI user metrics

  • Offering: GitLab.com, GitLab Self-Managed, GitLab Dedicated

The AiUserMetrics endpoint provides pre-aggregated per-user metrics for all registered GitLab Duo features, including Code Suggestions, GitLab Duo Chat, Code Review, Agent Platform, Job Troubleshooting, and Model Context Protocol (MCP) tool calls.

You can use this endpoint to analyze GitLab Duo user engagement and measure usage frequency across different GitLab Duo features.

Prerequisites:

  • You must have ClickHouse configured.

Total event counts

The AiUserMetrics endpoint provides the following levels of event count aggregation:

  • Top-level totalEventCount: Returns the sum of all event counts across all GitLab Duo features for a user.
  • Feature-level totalEventCount: Available in each feature metric type, returns the sum of all event counts for that specific feature.

You can use these fields to get aggregate counts at different levels of granularity.

For example, to retrieve both top-level and feature-level totals:

graphql
query {
  group(fullPath:"gitlab-org") {
    aiUserMetrics {
      nodes {
        user {
          username
        }
        totalEventCount
        codeSuggestions {
          totalEventCount
          codeSuggestionAcceptedInIdeEventCount
          codeSuggestionShownInIdeEventCount
        }
        chat {
          totalEventCount
          requestDuoChatResponseEventCount
        }
      }
    }
  }
}

The query returns the following output:

graphql
{
  "data": {
    "group": {
      "aiUserMetrics": {
        "nodes": [
          {
            "user": {
              "username": "USER_1"
            },
            "totalEventCount": 82,
            "codeSuggestions": {
              "totalEventCount": 60,
              "codeSuggestionAcceptedInIdeEventCount": 10,
              "codeSuggestionShownInIdeEventCount": 50
            },
            "chat": {
              "totalEventCount": 22,
              "requestDuoChatResponseEventCount": 22
            }
          },
          {
            "user": {
              "username": "USER_2"
            },
            "totalEventCount": 102,
            "codeSuggestions": {
              "totalEventCount": 72,
              "codeSuggestionAcceptedInIdeEventCount": 12,
              "codeSuggestionShownInIdeEventCount": 60
            },
            "chat": {
              "totalEventCount": 30,
              "requestDuoChatResponseEventCount": 30
            }
          }
        ]
      }
    }
  }
}

In this example:

  • The top-level totalEventCount (82 for USER_1) is the sum of all events across all features.
  • Each feature’s totalEventCount represents the sum of events within that feature only.
    • Code Suggestions: 60 events (10 accepted + 50 shown)
    • Chat: 22 events

Feature-specific metric types

The AiUserMetrics endpoint provides detailed metrics through feature-specific nested types. Each GitLab Duo feature has its own dedicated metric type that exposes event count fields for all tracked events related to that feature.

Available feature metric types include:

  • codeSuggestions: Code Suggestions-specific metrics
  • chat: GitLab Duo Chat-specific metrics
  • codeReview: Code Review-specific metrics
  • agentPlatform: Agent Platform-specific metrics (includes agentic Chat sessions)
  • troubleshootJob: Job troubleshooting-specific metrics
  • mcp: Model Context Protocol (MCP) tool call metrics

Each feature metric type includes:

  • Individual event count fields for all tracked events in that feature
  • A totalEventCount field that sums all events for that specific feature

The available event count fields are dynamically generated based on the events registered in the system. To see which events are tracked for each feature, examine the events declared in the ai_tracking.rb file.

For example, to retrieve detailed metrics across multiple GitLab Duo features:

graphql
query {
  group(fullPath:"gitlab-org") {
    aiUserMetrics {
      nodes {
        user {
          username
        }
        codeSuggestions {
          totalEventCount
          codeSuggestionAcceptedInIdeEventCount
          codeSuggestionShownInIdeEventCount
        }
        chat {
          totalEventCount
          requestDuoChatResponseEventCount
        }
        codeReview {
          totalEventCount
          requestReviewDuoCodeReviewOnMrByAuthorEventCount
          findNoIssuesDuoCodeReviewAfterReviewEventCount
        }
        agentPlatform {
          totalEventCount
          agentPlatformSessionStartedEventCount
          agentPlatformSessionFinishedEventCount
        }
      }
    }
  }
}

The query returns the following output:

graphql
{
  "data": {
    "group": {
      "aiUserMetrics": {
        "nodes": [
          {
            "user": {
              "username": "USER_1"
            },
            "codeSuggestions": {
              "totalEventCount": 60,
              "codeSuggestionAcceptedInIdeEventCount": 10,
              "codeSuggestionShownInIdeEventCount": 50
            },
            "chat": {
              "totalEventCount": 22,
              "requestDuoChatResponseEventCount": 22
            },
            "codeReview": {
              "totalEventCount": 8,
              "requestReviewDuoCodeReviewOnMrByAuthorEventCount": 5,
              "findNoIssuesDuoCodeReviewAfterReviewEventCount": 3
            },
            "agentPlatform": {
              "totalEventCount": 15,
              "agentPlatformSessionStartedEventCount": 8,
              "agentPlatformSessionFinishedEventCount": 7
            }
          },
          {
            "user": {
              "username": "USER_2"
            },
            "codeSuggestions": {
              "totalEventCount": 72,
              "codeSuggestionAcceptedInIdeEventCount": 12,
              "codeSuggestionShownInIdeEventCount": 60
            },
            "chat": {
              "totalEventCount": 30,
              "requestDuoChatResponseEventCount": 30
            },
            "codeReview": {
              "totalEventCount": 5,
              "requestReviewDuoCodeReviewOnMrByAuthorEventCount": 3,
              "findNoIssuesDuoCodeReviewAfterReviewEventCount": 2
            },
            "agentPlatform": {
              "totalEventCount": 20,
              "agentPlatformSessionStartedEventCount": 12,
              "agentPlatformSessionFinishedEventCount": 8
            }
          }
        ]
      }
    }
  }
}

Retrieve GitLab Duo and SDLC trend metrics

  • Offering: GitLab.com, GitLab Self-Managed, GitLab Dedicated

The AiMetrics endpoint powers the GitLab Duo and SDLC trends dashboard and provides the following pre-aggregated metrics for Code Suggestions and GitLab Duo Chat:

  • codeSuggestionsShown
  • codeSuggestionsAccepted
  • codeSuggestionAcceptanceRate
  • codeSuggestionUsers
  • duoChatUsers

Prerequisites:

  • You must have ClickHouse configured.

For example, to retrieve Code Suggestions and GitLab Duo Chat usage data for a specified time period for the gitlab-org group:

graphql
query {
  group(fullPath: "gitlab-org") {
    aiMetrics(startDate: "2024-12-01", endDate: "2024-12-31") {
      codeSuggestions{
        shownCount
        acceptedCount
        acceptedLinesOfCode
        shownLinesOfCode
      }
      codeContributorsCount
      duoChatContributorsCount
      duoUsedCount
    }
  }
}

The query returns the following output:

graphql
{
  "data": {
    "group": {
      "aiMetrics": {
        "codeSuggestions": {
          "shownCount": 88728,
          "acceptedCount": 7016,
          "acceptedLinesOfCode": 9334,
          "shownLinesOfCode": 124118
        },
        "codeContributorsCount": 719,
        "duoChatContributorsCount": 681,
        "duoUsedCount": 714
      }
    }
  },
}

Export AI metrics data to CSV

You can export AI metrics data to a CSV file with the GitLab AI Metrics Exporter tool.