infrastructure that watches itself so your team doesn't have to.

◉ the mechanism · what actually happens

One request, walked end to end — every step on your infrastructure, against your own models, with a human on the gate.

1. 01
   ChatMode

   You ask the state of the fleet in plain language — “anything drifting across the clusters?” ChatMode plans the query and dispatches its built-in agents — planning, data-query, validation, synthesis — to correlate signals across systems and surface what actually needs you. Not another dashboard to scan; the one that already read the others.

2. 02
   SmartModelRouter

   Every step routes to a model that fits it — a fast model to classify a flood of alerts, a long-context model to read a drift report, a strong reasoner to judge whether an anomaly is real or noise — across the providers you register. Bring your own models and run them on your own infrastructure; the telemetry, the topology, the config never leave your network.

3. 03
   Usage analytics

   Every model call the watch makes is metered — tokens, requests, and spend tracked over time, per provider and model — so the cost of the automation is never a mystery. You can see exactly what the watch is spending as the estate grows, and catch a runaway before the invoice does.

4. 04
   MCP tools · OBO credentials

   The watch reaches your estate as a fleet of MCP tools — across Kubernetes, AWS, Azure, GCP, Prometheus, and infra-health checks — every call running under your own identity, with your scoped credentials forwarded per call. List nodes, query metrics, run a connectivity or Redis health check, pull cost by service: least privilege by construction, not by policy memo, and every call logged.

5. 05
   AgenticWorkflows

   Self-healing operations run as flows of named agents: detect the anomaly, correlate the cause, check it against your compliance baseline, draft and stage the fix. The watch runs itself — and there’s more behind sign-up: drift and config-compliance sweeps, multi-cloud capacity forecasting, the auto-remediation runbooks that stage the rollback before they ever ask to run it.

6. 06
   HITL approval · DLP

   Any operation that changes live infrastructure stops at the human gate — self-healing, but never unsupervised. The gate is real architecture: any acting step pauses for a person, and if no one approves, it times out and is denied. DLP masks secrets before anything is stored or sent, so the telemetry that flows through the watch never leaks a credential.

7. 07
   Code Mode · build the runbook tool, don’t wait for it

   The remediation tool you keep meaning to write between pages — Code Mode drafts it against your stack and proves it with tests, gated by your review before anything touches an environment. You stay on the incident and the judgment calls; the agent builds the automation you’d otherwise queue for a quarter, so the on-call rotation you have covers more ground.

8. 08
   Audit trail · RBAC

   Every detection and every action is written to an append-only audit log — once recorded, frozen, so the trail is tamper-evident — and role-based access scopes who can let an autonomous operation act. Full accountability for self-healing: you can show exactly what the watch saw, what it proposed, who approved it, and what it did.

The infrastructure watches itself, the fixes wait for your nod, and the on-call sleeps — and that’s the cluster watch. The drift sweeps, the compliance checks, and the multi-cloud remediation are waiting behind sign-up.