An audit trail for AI coding agents
When an AI agent changes your code, you should be able to answer later who asked for the change, what the agent planned, who approved it, what ran, and what passed. Dualzo records every step in an append-only audit log, from the GitHub issue to the merged pull request.
What is recorded
| Step | What the log keeps |
|---|---|
| Intake | The GitHub webhook or prompt that started the task, including webhooks Dualzo chose to ignore. |
| Plan | When the plan was built, and when it was rebuilt with new instructions. |
| Approval | Who approved, modified, or rejected the plan, and when — including auto-approved runs. |
| Execution | When the run starts, and each task as it passes or fails. |
| Validation | Each fix attempt after a failing check. The commands and output of every validation run are kept on the task page. |
| Pull request | The pull request being opened, then merged or closed on GitHub. |
| Workspace | Repository connections, settings changes, and billing changes. Saved test variables are logged by count, never by value. |
Each entry carries the actor — the person who acted, or the system — and the time, in UTC.
Append-only by design
There is no way to edit or delete an audit entry in Dualzo — not for members, not for admins. Entries are kept for 90 days and then pruned automatically. Each workspace's log is isolated: no one outside the workspace can read it.
Reading the log
- Workspace audit page — the whole workspace's log, filterable by event, actor, and date range.
- Task timeline — every event for a single task, in order, from intake to pull request.
- From the pull request — every pull request Dualzo opens links back to that task's audit page, so a reviewer on GitHub is one click from the full history.
Why an audit trail matters for AI-written code
Git history shows what changed. It does not show that a person approved the plan before the change was written, which checks ran, or which attempts failed along the way. When a team lead, a security reviewer, or an auditor asks how AI-written code reached production, the audit trail is the answer — and it is written as the work happens, not reconstructed afterwards.
The audit trail is one of the four controls of AI coding governance, together with a written plan, human approval, and sandboxed validation.
Team plan — $119/month, 5 seats included. AI model usage included.