It acts alone.
Only inside your rules.
Most platforms promise safety in a whitepaper. Cue makes it a surface you can open: named checkpoints, per-agent scopes and budgets, policy-aware model routing, and a live ledger of every act and every dollar.
Built in, not bolted on.
"Never send email without confirmation." "Anything customer-facing waits for me." Rules written in plain English, enforced on every agent, firing visibly when they catch something.
Every agent has a ceiling. When Ops hits its $2K cap it pauses itself and tells you why — raise it and continue, or leave it parked. Overruns are impossible by construction.
Route each task by cost, accuracy and policy. Pin a model to an agent, keep sensitive work on approved models, and see which model did what — no lock-in, ever.
Every act every agent takes, timestamped and attributed — with live spend per agent, per mission, per model. Not a monthly invoice surprise: a running meter you can open any time.
Agents earn autonomy.
You grant it.
New agents start supervised. As a track record builds — acts taken, zero reversed — they request wider scopes through a proper ceremony: what they want to do alone, the evidence, and a one-tap grant you can revoke any time.
The Guardrails console.
Checkpoints, agent scopes, spend and the ledger — one place to set your rules and watch them hold.
