Infrastructure
A root cluster provisions and manages the lifecycle of a fleet of purpose-built child clusters.
Platform
Opinionated, integrated, self-operating.
Pax Agentica is a deliberate selection of well-established open-source components, best practices and agentic AI — configured, integrated and operated as one product. It is shaped like a multi-tenant SaaS system, where each tenant is a department, business unit or team in your organisation.
Three layers
Each layer has its own lifecycle and blast radius, and every architectural decision — identity, networking, data isolation, observability, quota and billing — follows standard multi-tenant patterns.
Layer 1
Layer 2
Platform
Dedicated clusters for operations, observability, security, developer tools, inference, messaging, data, agents, applications, testing and edge.
Layer 3
Application
An event-driven framework (CQRS and domain-driven design) for building business domains: write APIs validate and publish events; read models serve queries.
The cluster fleet
Every cluster shares a common baseline — certificate management, service mesh, event-driven autoscaling and telemetry collection.
| Cluster | Purpose | Core components |
|---|---|---|
| Root | Fleet provisioning and lifecycle | Cluster API, k3k |
| Operations | Pipelines, artefacts, backup | Tekton, MinIO, registry, Velero |
| Observation | Metrics, logs, traces, alerting | SigNoz, ClickHouse, OpenTelemetry |
| Security | Identity, secrets, policy, runtime security, supply chain | OpenBao, SPIFFE/SPIRE, Keycloak, Kyverno, Falco, Trivy, Cosign |
| Dev tools | Source control, planning, code quality | Gitea, OpenProject, SonarQube |
| Inference | Self-hosted model serving | vLLM, KubeAI |
| Bus | Event backbone and schemas | Redpanda, Schema Registry |
| Database | Relational, document, graph, vector, cache | PostgreSQL, MongoDB, Neo4j, pgvector, Redis |
| Agents & MCP | Agent runtimes and tool servers | Agent services, MCP servers |
| Apps | Your business domains | CQRS and DDD domain services |
| Tests | Load, security, chaos, end-to-end, contract | k6, Trivy, kube-hunter, Chaos Mesh, Playwright, Pact |
| Edge | Ingress, egress, WAF, authorisation | Istio, external-dns, Coraza WAF |
Security model
Designed for the scrutiny of central-bank, insurance and telecoms regulators.
- Workload identity everywhereEvery service and agent receives a cryptographic identity. People and agents get short-lived tokens scoped to a single audience.
- Secrets that expire on useSecrets are issued once and delivered in memory. Instead of rotating credentials, the platform replaces the workload.
- Agents in a sandboxAgent tool execution is isolated, network egress is allowlisted, and authorisation is bound to a specific work package.
- Signed supply chainArtefacts are scanned, signed and verified before they are admitted to a cluster; policy is enforced at admission.
- Tenant isolationBusiness units are isolated across identity, network and data, with quota and cost attributed per tenant.
- Full audit trailEvery change is a commit and every action an event — giving auditors a single, complete record.
What Pax Agentica is not.
Being clear about scope is part of the design.
Not a generic distribution
Fixed components, real integration
The component choices are deliberate. The value lies in how they are integrated and operated, not in endless configurability.
Not a hosted service
It runs inside your boundary
Pax Agentica runs in your accounts. Operating it as a managed service would mean giving up the sovereignty it exists to protect.
Not AI for everything
Agents build classical software
Agents create and run conventional software, because that is faster and cheaper than using AI for every task. Over time, more of it takes the form of MCP tools that agents orchestrate.
Not tied to one cloud
Open standards over proprietary services
Where an open alternative exists, we use it. Today that means a minimal AWS footprint, with Azure, GCP, Huawei Cloud and others to follow.
Next step
See how Pax Agentica would run inside your boundaries.
A briefing walks through the platform, the control model and an adoption path that fits your regulators, your teams and your cloud.