Agentic AI under your control.
An AI employee that plans and calls your tools behind an allowlist — with handoff and human approval for sensitive writes.
Real governance
Per-tool policy
off / suggest / auto per organization.
v1 tools
Knowledge, tags, handoff, ticket, store order, webhook, enqueue Flow.
Inbox approvals
Sensitive tools wait for an agent before executing.
Containment
Measure no-human resolution, tool success, forced handoff.
Billed to AI quota
Agent steps count against the existing ai_replies_limit.
Inside Flow Studio
agentic_step uses the same runtime.
FAQ
No — every tool passes PolicyGuard; sensitive tools start in suggest mode.
Existing AI allowance (typically Business ai_full) — no separate meter at launch.