Open-Source · Self-Hosted · AI-Ready
Most infrastructure was designed for humans with shell access and tribal knowledge. The next era needs something different — explicit state, typed operations, and infrastructure that AI agents can safely understand and act on. Globular is built for that shift.
The Problem
Shell access is the most dangerous control surface you can give an AI agent. Logs are too unstructured to reason from. Dashboards don't expose typed operations. For AI to participate in operations safely, the platform must expose structured, verifiable, auditable operational truth at every layer.
Four Truth Layers
Globular separates cluster state into four independent layers, each with its own owner and data source. They are never collapsed. The gap between layers is exactly where problems live — and where AI agents and operators look first.
Does this version exist? Artifacts are published, checksummed, and versioned in MinIO-backed object storage.
What should be running? The controller writes desired service versions to etcd — the single source of intent.
What is actually installed? Node agents report installed packages with checksums. Drift from desired state is visible immediately.
Is it running and healthy? systemd reports unit state. The cluster doctor correlates all four layers and surfaces typed findings.
The AI Operation Plane
Globular does not give AI a dangerous shell. It gives AI a secured, discoverable, typed, permissioned control surface. Every action is RBAC-controlled. Mutating operations require workflows. The cluster doctor verifies every result.
AI agents read real cluster state, installed packages, doctor findings, workflow history, and drift reports through structured APIs — no log scraping.
Tier-1 actions run automatically. Tier-2 actions require human approval. Mutating operations flow through idempotent workflows, never raw commands.
Doctor findings and convergence checks confirm the cluster reached the desired state. Every AI action is recorded in the workflow audit trail.
Positioning
Kubernetes orchestrates containers. Globular is the substrate layer that was never named: cluster identity, PKI, service versioning, workflow execution, health convergence, and AI-readable operational truth. Globular can complement Kubernetes — or serve workloads that do not need container orchestration at all.
| Traditional platforms | Globular | |
|---|---|---|
| State model | Implicit, eventually consistent | Explicit four-layer truth model |
| AI control surface | Shell access, log scraping | gRPC + MCP, RBAC-controlled |
| Operations | Shell scripts, tribal knowledge | Typed workflows, idempotent, audited |
| Diagnosis | Human log review | Cluster Doctor with structured findings |
| Identity & PKI | External tooling required | Built-in mTLS, CA, JWT, RBAC |
| Service discovery | DNS or sidecar proxy | etcd-backed, no sidecars |
| Runtime | Containers required | Native Linux binaries under systemd |
| AI audit trail | None | Every action recorded in workflow history |
Globular runs in production on a 3-node bare-metal cluster, operated daily with Claude through MCP. Every design decision, failure, and recovery is documented and open.