Get AI commit details with the AI Code Tracking API

Retrieve blame annotations and referenced conversation metadata for one or more commits with GET https://api.cursor.com/analytics/ai-code/commits/:commitHash. Official reference: AI Code Tracking API → Get Commit Details.

This endpoint is in limited alpha and available only to select users. Response shapes may change. Authenticate with a team API key (Basic auth, key as username, empty password — same method as the Admin API). Create keys at cursor.com/dashboard → API Keys with scope admin:*. Broader AI Code Tracking availability is Enterprise only (contact sales). Rate limit: 20 requests per minute per endpoint. Metrics cover the git repository at the top level of the workspace root; multi-root workspaces are unsupported.

Path and query

Field	Rules
commitHash (path)	Single hash, or comma-separated list (abc123,def456)
branch (query)	Optional branch name filter

Response shape is the same for one hash or many: an object with commits and conversations arrays.

Single commit

curl -X GET "https://api.cursor.com/analytics/ai-code/commits/0aabf603dc906e05bf5e4d9fd423fdd517f2e43f?branch=main" \
  -u YOUR_API_KEY:

Multiple commits

curl -X GET "https://api.cursor.com/analytics/ai-code/commits/abc123,def456,ghi789" \
  -u YOUR_API_KEY:

Example response:

{
  "commits": [
    {
      "commitHash": "0aabf603dc906e05bf5e4d9fd423fdd517f2e43f",
      "commitSource": "ide",
      "rangeAnnotations": [
        {
          "filePath": "src/analytics/report.ts",
          "groups": [
            {
              "conversationId": "conv_abc123",
              "model": "gpt-4o",
              "operationType": "insert",
              "ranges": [
                { "start": 10, "end": 25 },
                { "start": 42, "end": 58 }
              ]
            }
          ]
        }
      ]
    }
  ],
  "conversations": [
    {
      "id": "conv_abc123",
      "title": "Refactor analytics module",
      "tldr": "Extracted report generation into separate functions",
      "overview": "Refactored the analytics module to improve maintainability by extracting report generation logic.",
      "summaryBullets": [
        "Created dedicated report generator class",
        "Added unit tests for new functions",
        "Updated imports across affected files"
      ]
    }
  ]
}

Response fields

Field	Type	Meaning
commits	array	Commit objects with blame annotations
commits[].commitSource	"ide" | "cli" | "cloud"	Where the commit originated
commits[].rangeAnnotations	array	File-level blame data
commits[].rangeAnnotations[].filePath	string	Path inside the repo
commits[].rangeAnnotations[].groups	array	Annotation groups
groups[].conversationId	string | null	Conversation that generated the code
groups[].model	string | null	Model used
groups[].operationType	string	Operation type (for example insert)
groups[].ranges	array	Line ranges (start, end)
conversations	array	Metadata for every referenced conversation
conversations[].id	string	Conversation id (matches conversationId on groups)
conversations[].title / tldr / overview	string | null	Title and summaries
conversations[].summaryBullets	array | null	Bullet summary points

Join with usage events

conversationId on annotation groups is the Agent session id. Join it with conversationId on rows from Get filtered usage events when you need cost / token rows for the same session.

Pitfalls

• Calling without limited-alpha access — expect failures until Cursor enables the route for your team.
• Treating the response as a single commit object — even one hash returns { commits, conversations }.
• Assuming every group has a conversationId — the field can be null.
• Multi-root workspaces — only the top-level git repo is covered.
• Alpha churn on blame / conversation fields — pin docs before automating.

Aggregated commit lines: Get AI commit metrics. Accepted changes without a commit: Get AI code change metrics.