Verify deployment

Use this guide to confirm AI Code Insights is installed correctly and data is reaching DX. You’ll check daemon health, identity, repository matching, agent integrations, and attribution.

Before you begin

  • A supported machine: Run the daemon on the machine where the repository and coding agent run.
  • A monitored repository: Import the repository into DX and confirm its local origin can be matched to the canonical URL.
  • A supported coding agent: Use a tool listed in Supported coding agents.
  • Report access: Ask the tester to open their Personal dashboard. For attribution checks, make sure someone with drilldown access can open commit details in AI code percentage.

Step 1: Check daemon health

Run the diagnostic command on the developer machine.

  1. macOS and Linux
  2. Windows
aicodemetricsd status
& "$env:ProgramFiles\AICodeMetrics\aicodemetrics.cmd" status

Resolve every required [FAIL]. Review [WARN] entries for missing hooks, identity, protected-directory access, or repository coverage before continuing.

Step 2: Confirm identity and repository matching

In the status output, verify that:

  • The configured email matches the developer’s DX email.
  • API connectivity succeeds.
  • The test repository appears under monitored repositories.
  • The intended coding agent is detected and its integration is installed.

If the repository is missing, compare git remote -v with the repository URL in DX. Configure url_rewrites when an internal mirror or SSH alias changes the URL.

Step 3: Create a known test change

Have the tester perform this step on their machine in the monitored repository.

  1. Start a new session with the supported agent inside the monitored repository.
  2. Ask the agent to make a small, recognizable code change.
  3. Make a small manual edit in the same change so the commit contains both known AI and human input.
  4. Commit and push the change to a remote branch.
  5. End the agent session if you are also validating session data.

Commit attribution is uploaded after the daemon detects the commit on a remote branch. Session data is uploaded when the integration emits a stop, compact, or end event.

Starting the session from the repository’s parent directory does not enable code attribution, even when the agent edits files inside the monitored repository. Additional session directories expand session collection only.

Step 4: Inspect the commit in DX

Ask the tester to open their Personal dashboard and confirm they can see their own commit activity for the test change.

Use this as a visibility check only. To complete daemon attribution validation, open AI code percentage, select the test period and repository, and open the drilldown for the test commit. If drilldown access is restricted, ask an admin to run this check.

Note: If individual data access is disabled enterprise-wide, non-admin users cannot open commit drilldowns in AI code percentage.

In the drilldown, confirm that:

  • The commit appears under the expected developer and repository.
  • The AI code percentage is directionally consistent with the known change.
  • The AI tool and daemon version match the test machine when those fields are available.

If the expected percentage differs materially, hover over the commit row, click the pencil icon next to AI code %, enter the correct percentage from 0 to 100, and click Save. Developers can edit their own linked commits; workspace admins and Data Cloud admins can edit any commit they can access. The saved override is marked Edited and is used in the report’s weighted calculations.

Record the original percentage and commit SHA before making a correction so you can still assess the daemon’s measured attribution. For help investigating a discrepancy, contact DX support with those details, the expected percentage, and relevant workflow context.

Step 5: Validate session data

When Transcripts & Agent Experience is enabled, open Agent Experience Score after the session finishes.

Confirm that the session appears under the expected developer and tool. A score may remain unavailable when the session is too long or contains insufficient evidence for evaluation.

Next steps