Security
What is Approval Tiers?
Also known as: tiered tools, action tiers
Approval tiers classify an agent's actions by how much damage they can do — typically read, write and destructive — and require human sign-off only on the tiers where a mistake actually matters.
Approving everything trains people to click yes without reading; approving nothing is how an agent empties a folder. Tiering resolves that by grading actions on reversibility. Reading a calendar is a read: wrong is harmless. Drafting a message is a write: wrong is embarrassing but fixable. Deleting a repo, sending to a client list, or paying an invoice is destructive: wrong is permanent. Sign-off is spent where it buys something, so the prompts you do see are rare enough to be worth reading.
This is how tools work in Velaris: every tool in the capability mesh is tagged read, write or destructive, and destructive-tier calls stop the run and wait for a human decision. Because agents are multi-step graphs, the pause is a real suspend — the run holds its state, you see the exact action proposed, and it resumes from that point on approval rather than restarting. Combined with per-run scoped credentials, the tier decides what needs your attention and the token decides what's even reachable.
Related terms
See it in action
Velaris puts approval tiers to work inside a real AI Operating System.
Get early access