Our principles
What Harness will always do — and what it never will.
Always
- Every agent action is logged before it happens, not after.
- Risky actions — money, deletions, outbound mail — wait for a human yes.
- Access is granted per integration, and can be revoked in one click.
- You can export your full audit history at any time.
Never
- We never train models on your tasks, documents or logs.
- We never let an agent widen its own permissions.
- We never hide a failure — a stuck or failed task is loud, not silent.
- We never sell or share your data. Your operations are not a product.