DocsUsing Grapim
Reports
How much code agents write, what reached the default branch without review, and how each agent performs.
Reports are on the Governance plan. Teams on Free and Team see a preview with last week's AI share. There is no checkout during early access; Governance is available as a pilot, so talk to us to try it.
The reports
| Report | What it answers |
|---|---|
| Merged without review | AI-written lines that reached the default branch from runs nobody approved: pending, rejected or never reviewed, by run, repo and agent. |
| AI share | The share of added lines that agents wrote, by week, repo and agent. |
| Agents | Per agent: runs, lines, undo rate, reject rate and median review time. |
| Activity | Runs per day, files touched, edits that were blocked, and files two runs changed on the same day. |
Pick a range (4 weeks, 12 weeks, 6 months, a year, or exact dates) and a repo. The range is limited to your plan's history: 365 days on Governance.
Export
Every report has Export CSV, with the same range and repo. Cells that a spreadsheet would read as a formula are made inert.
Audit evidence
Reports, Audit evidence lists every agent run synced in a date range, for an auditor who samples changes:
- the repo, the agent, what it was asked (the first prompt), lines added and removed, and who synced it;
- every decision in the hub: approved or rejected, by whom and when, and whether it was made from the GitHub check;
- whether the run reached a merged pull request, with its link (pull requests the GitHub App checked and saw merged);
- the path rules that apply to it now, and those in force at each decision.
Download it as CSV, or open the Print view and save it as PDF. Both say what they are: Grapim's record of runs synced to it and decisions made in it, evidence for your auditor to evaluate. It is not a certification, and runs on machines without Grapim are not in it.
How the numbers are counted
- AI lines come from the provenance notes; human lines are the rest of the lines in recorded commits.
- Commits without a note were not recorded. Their lines are unknown: shown as a coverage gap, never counted as human. Each report says how many repos and commits it covers.
- Lines from undo, restore and folder-sync runs are people's work.
- The review state of a run is its newest decision in the hub.
Reports are cached for up to an hour, and refreshed as soon as a machine syncs.