KOSMOY MISSION CONTROL

Govern agents as operational systems, not just prompts.

An agent is more than a prompt. It has identity, permissions, tools, memory, model access, runtime environment, an oversight path, and an evidence trail. Mission Control gives platform teams and operators a fleet view of all of them.

Agent fleet · 6 agents · last 24h

live

AgentOwnerStatusUse caseCost (24h)Last runRisk
Sales-support assistantCommercialRunningSales-support — copper (EU)$42.182m ago
Legal contract reviewLegal ITRunningContract review (EU)$11.0418m ago
HR engagement botPeople OpsIdleEmployee engagement$0.841h ago1 PII
Market intelligenceStrategyStoppedCompetitor monitoring (DACH)$012m agoKILL SWITCH
M365 productivityWorkplaceRunningProductivity assistant$128.434s ago
Developer GatewayPlatform ITRunningCoding agents (Claude · Cursor)$76.911s ago
Six large cards show a supervised agent task moving from the human to Kosmoy Chat, Mission Control, Action Plane, Action Capsule and approved egress. Evidence flows back to the AI system dossier.Mission Control · Agent Supervision PathEvery agent task. Supervised from start to finish.1HumanSubmits task,approves, or stops.2Kosmoy ChatThreads, result cards,clarifications.3Mission ControlIdentity, runtime status,run history, kill switch.4Action PlanePre-flight checks,JIT credentials issued.5Action CapsuleSandboxed run,run-scoped lease.6Approved egressThrough paired Gateway.All logged.Evidence flows backAI system dossier / Insights DashboardComplete audit trail, citations, logs and metrics.

Identity, permissions, oversight, stop, evidence.

Identity

Each agent has a runtime identity, scoped permissions, and a short-lived credential profile that flows from Mission Control to the Action Plane to the Capsule.

Permissions

What models, MCP servers, A2A peers and HTTPS endpoints the agent can reach. Bounded by the paired Gateway. Governed by use case.

Oversight

Human-in-the-loop primitives: clarification, approval, stop, rerun. All surfaced in Kosmoy Chat, all logged.

Stop controls

Pause schedules. Cancel future work. Deny new Action Plane calls. Kill switch on a live run.

Evidence

Every run links back to the approved AI use case. Tasks, runs, status changes, outputs, policy events, cost — searchable in Insights and bound to the dossier.


Module questions, answered straight.

Is Mission Control the same as the Insights Dashboard?

No. Insights is the cross-app analytics view (cost / quality / risk signals). Mission Control is the agent-fleet operational view: identity, runtime status, oversight actions, kill switch.

What's the difference between Mission Control and the Action Plane?

Mission Control is the human-facing supervision surface. The Action Plane is the runtime control plane that admits, authorizes, and audits each agent run before it executes inside a Capsule.

Does Mission Control work for agents on Foundry or Bedrock?

Inventory and status are reflected for external agents (where the upstream platform exposes it). Runtime control — kill switch, pre-flight authorization, JIT credentials — only applies to agents running in a Kosmoy Action Capsule.

Can a non-operator approve an agent action from Mission Control?

Yes. Approvals are role-scoped. End users approve their own task's bounded actions in Kosmoy Chat; operators approve in Mission Control. Both produce the same audit record.

See Mission Control in production.

Walk through the fleet view, an oversight action, and the kill switch.