An operator that actually does the work.
Give it a goal — process refunds, resolve a damaged order, reconcile invoices. It plans, gathers facts, follows policy, and executes across real tools. Anything that moves money or sends mail stops for your one-tap approval, and every step is logged.
What this demo proves I can build
◆Autonomous planning
Turns a plain-English goal into a concrete plan, then works the plan step by step.
⚙Real tool use
Reads & writes tickets, orders, invoices, stock and email via live function-calling.
⚑Human-in-the-loop
Stops for one-tap approval before anything that moves money or sends mail.
🛡Policy guardrails
Follows a written policy; unsafe actions are hard-blocked, not just discouraged.
⟲Durable & auditable
Every step is logged and replayable — reopen a run and it resumes after a restart.
◷Bounded by budgets
Hard caps on steps and tokens keep each run predictable and cost-safe.
Starting the live demo…