GVNR
Spend caps are not unique and it would be dishonest to imply otherwise. Provider limits cap usage at organisation, workspace and member level. LLM gateways and hosted spend caps stop spend before the call. Observability tools report what was spent.
Two kinds of tool get close, and each covers half. Gateways and spend caps stop the money but have no opinion on what the agent is doing. Command guards block the dangerous command but never look at the money. GVNR does both on the same action, before it runs, and keeps a receipt of every decision naming the person the agent acted for.
| Stops spend | Limits the rate | Gates the action | Team policy | Record | |
|---|---|---|---|---|---|
| Provider limits | at a set total | no | no | member limits | usage reports |
| Observability tools | no, they report | no | no | no | logs |
| LLM gateways | yes, per key | configurable | no | per key | logs |
| Hosted spend caps | yes | some | no | varies | varies |
| Command guards | no | no | yes, before it runs | rarely | not their focus |
| GVNR | yes, per agent and in total | yes, on by default | yes, before it runs | optional, Rego in your workspace | hash-chained receipts, local and shared |
A spend cap answers "can it afford this?". It has no opinion on curl | sh, on rm -rf, or on reading your .env, all of which are cheap. A command guard answers "is this command dangerous?" and never notices a session fanning out into subagents at dollars a minute. GVNR asks both, on the same action, before it runs, and where a safer form exists it offers that instead of refusing.
It does not detect hallucination. Nobody can do that reliably. It catches mechanical waste that is genuinely detectable, blocks capability misuse, and asks you about the judgment calls.