Governance OS for AI agents
Permission for your agents. Control for you.
Human approval infrastructure for autonomous systems. Agents request; humans decide; every action is logged, signed, and human-gated (or autonomous) — with budgets, org structure, and a bring-your-own-key LLM gateway built in.
Logged · signed · human-gated · self-hostable · BYOK
CriticalPendingSpendfrom infra-bot
Approve $12,000 cloud overage
This month is 4× budget. Approve the overage or cap spend.
1 Approve2 Reject4 Escalate
9
card types
BYOK
LLM gateway
100%
append-only audit
MCP + REST
agent-native
Everything you need to govern AI agents
From a single approval card to a fleet of budgeted, orchestrated agents.
◆
Approval cards
Nine risk-ranked card types — approval, spend, info, escalation, polls, choices. Expiring and re-armable.
◆
Process-cards mode
Clear the queue one card at a time with arm-then-confirm actions and 1–5 hotkeys.
◆
Agent org chart
Who owns each agent, the orchestration hierarchy, who's accountable — visualized with budgets.
◆
Token budgets
Meter burn per agent and team, track cost vs budget, flag overruns in red.
◆
BYOK LLM gateway
Vendor-agnostic proxy: your keys, scoped issued keys, network/rate/size/budget guardrails.
◆
Append-only audit
HMAC-signed webhooks or long-poll. Every action recorded — nothing mutated or deleted.
How it works
1
The agent asks
Your agent submits a card over MCP or REST — with risk, routing, and budget context.
2
A human decides
It lands in the inbox. Approve, reject, request info, escalate, or bounce it back with edits.
3
Logged & auditable
The decision is signed, returned over webhook/long-poll, and written to an append-only audit.
Give your agents permission — keep control.
Set up in minutes. Bring your own keys. Every decision logged, signed, and human-gated (or autonomous).