Situation
A retail giant ran customer-facing agents across regions, brands, and vendors: OpenAI here, Anthropic there, custom gateways everywhere. Leadership could not answer 'how many agents do we have?' let alone 'what happens if we need to stop them all?'
What was at stake
Black Friday traffic risked runaway token spend and inconsistent safety policies. A single compromised agent with broad tool access could touch loyalty PII at scale.
What Threesixty did
Inventory and migrate agents to gateway-routed hosts with Command Center as system of record.
Deployed global safety policies with fleet-wide apply and per-brand overrides where required.
Set per-customer and per-workload token budgets with alerts before spend surprises hit finance.
Exposed executive dashboard and optional client portal views for SLA and incident transparency.
Technical approach
Command Center fleet operations: multi-agent gateway per machine, per-customer token budgets, provider credential management with rotation visibility. ClawGuard policies with auto-apply to new machine registrations. White-label portal options for B2B partners; staged releases via platform ship-confidence pipeline for prompt and model changes.
Results
- Over 1,200 agents under single fleet visibility with global kill-switch tested quarterly.
- Token overspend incidents down sixty-eight percent year-over-year with budget alerts.
- Fleet-wide policy changes propagated in under four hours, not weeks of regional tickets.
- CX and security leadership share one source of truth for violations, spend, and uptime.