Silicon Cortex: The Thermal Pattern
Event-driven AI agent orchestration on Kubernetes. Zero idle cost. Every task spawns, executes, and terminates. Nothing wasted.
Silicon Cortex
Most agent platforms keep workers idling, burning cost while waiting for work. Silicon Cortex flips the model. An event-driven AI software engineering platform on Kubernetes that spawns ephemeral agent pods on-demand, generates production code with Claude, pushes it to GitHub, and prints a physical receipt when it's done.
Knative Eventing with CloudEvents for loose coupling. Python FastAPI services for task control and agent spawning. Redis for state and queues. Claude Code CLI running inside ephemeral Kubernetes Jobs that create repos, write code, and terminate. Nginx gateway routing traffic to all services. A NIIMBOT B1 thermal printer producing physical receipts for every completed task.
“Thermal, not inkjet. Heat and paper - nothing wasted when idle.”
Key Integrations
The Thermal Pattern
How It Works
Why Thermal?
No agent pods run when there's no work. Pay only for compute you actually use.
Services communicate via CloudEvents through a Knative Broker. No direct calls, no polling, no wasted cycles.
Each task gets exactly one K8s Job. It runs, completes, and cleans up. Like thermal printing: heat and paper, nothing wasted.
Components
Eight services communicating through CloudEvents. Each does one thing well.
Task Controller
Manages task lifecycle, stores state in Redis, emits CloudEvents
Python + FastAPIAgent Spawner
Creates ephemeral K8s Jobs on-demand when tasks arrive
Python + FastAPIAgent Pod
Runs Claude Code CLI to generate code, push to GitHub, then terminates
Node.js + Claude CodePrint Queue
Queues physical receipt jobs on task completion via thermal printer
Python + NIIMBOTArchitecture
Nginx gateway, FastAPI services, Knative event mesh, ephemeral K8s Jobs, Redis state, and a thermal printer.
Interactive diagram powered by ArchByte - pan, zoom, click nodes, and explore flows