See the gate in action

A real run captured from Ruligent's internal demo agent on August 4, 2026 — not a mock-up.

Below, a sales agent makes six tool calls through the Ruligent guard. One sails through, one hits a destructive-operation policy, one waits for a human, one spends within budget, one blows the budget, and then the kill switch ends the party. Every decision lands in the audit log.

=== Ruligent demo: sales-agent-01 goes to work ===

✔ ALLOWED  search_crm — No matching policy; default is allow
✘ BLOCKED  delete_contact — Blocked by rule "Block destructive operations"
⏳ PENDING  send_email — waiting for a human in the approval inbox...
   (reviewer clicked Approve in the inbox)
✔ APPROVED send_email — Approved by www.ruligent.com/contact
✔ ALLOWED  llm_completion ($20) — within the $25/day budget
✘ LIMITED  llm_completion ($10) — Agent "sales-agent-01" spend limit of $25/day exceeded ($20 spent, $10 requested)
   (admin flips the org-wide kill switch)
✘ KILLED   search_crm — Kill switch is active for the entire organization

=== Audit log (6 guarded calls) ===
2026-08-04T19:29:43.915Z  search_crm       allow             risk=low
2026-08-04T19:29:43.919Z  delete_contact   block             risk=critical
2026-08-04T19:29:43.923Z  send_email       require_approval  risk=high
2026-08-04T19:29:44.345Z  llm_completion   allow             risk=medium
2026-08-04T19:29:44.352Z  llm_completion   spend_limited     risk=medium
2026-08-04T19:29:44.361Z  search_crm       killed            risk=low

Demo complete. Every one of those decisions is queryable, exportable, and webhook-notified.

What just happened

Default allow, fully logged — search_crm matched no policy, so it ran; the decision is still recorded with a risk label.
Policies block the scary stuff — delete_contact hit the "Block destructive operations" rule and never executed.
Humans stay in the loop — send_email paused in the approval inbox until a reviewer approved it; the approval records who and when.
Spend limits hold the line — the first llm_completion fit the $25/day budget, the second was refused mid-stride.
The kill switch beats everything — once flipped, even previously allowed calls are stopped.

Try it in the hosted product

Create a free account, connect an agent through the REST API, and reproduce the same allow, block, approval, spend-limit, and kill-switch sequence against non-production test tools.

Use the live API quickstart or book a walkthrough.

Next steps

Start free Read the docs Book a walkthrough