Contact Sales →
Platform Architecture

The Tech stack, layer by layer.

A request enters at the edge, runs through an engine that executes code or model weights, then reads or writes data. Here is the platform that makes that happen.

Home/Platform
The unifying model

Node runs the assistant, Lucee → Tomcat → JVM runs the web apps, and Ollama runs the models — while nginx is the shared front door to all three.

The Stack

Eight layers, each with one job.

From bare metal up to the deploy pipeline — the vocabulary anchored to where it actually lives.

01 · INFRASTRUCTURE

Hosting

  • IONOS bare-metal IaaS
  • VPS fleet + a dedicated GPU box
  • Ubuntu 24.04 LTS throughout
02 · PROCESS

Containers

  • Docker images & Compose
  • systemd keeps services alive
  • Bind mounts persist state
03 · EDGE

Network

  • nginx reverse proxy + TLS
  • Let's Encrypt auto-renewal
  • Loopback binding & auth gates
04 · ENGINES

Execution

  • Node.js for the assistant
  • Ollama for inference
  • Lucee → Tomcat → JVM for CFML
05 · APPLICATION

Apps

  • OpenClaw / MAX assistant
  • CFML sites: isonar, legalseer…
  • Mission Control dashboard
06 · AI

Inference

  • Models & weights (gemma, OCR)
  • GPU1 as the inference server
  • Endpoints, context window, RAG
07 · DATA

Storage

  • PostgreSQL 17 system of record
  • Lucee datasources + JDBC
  • SQLite for Mission Control
08 · DEPLOY

Source & CI/CD

  • Git forks, main vs. PROD
  • GitHub Actions pipeline
  • rsync deploy + build artifacts
★ FRONT DOOR

nginx ties it together

  • One edge in front of all three engines
  • PostgreSQL is the system of record
Layer 4 · Engines

Three engines, one edge.

An engine is software that takes code or model weights and actually executes them. The platform runs three, each sitting behind the same nginx front door.

  • JS
    Node.jsThe JavaScript runtime executing the OpenClaw gateway.
  • AI
    OllamaThe inference engine that loads and runs model weights.
  • CF
    Lucee → Tomcat → JVMCFML compiled to Java bytecode, served as servlets on the JVM.
Request flow · one isonar page
1 nginx receives HTTPS
2 Forwards to Tomcat
3 Tomcat routes to Lucee (a servlet)
4 Lucee runs compiled CFML on the JVM
5 Response flows back out
If you keep four sentences

The quick mental map.

GPU1 · '134

The inference server. Hosts the models — Ollama is the engine; gemma and the OCR models are the weights.

MAX · '16

The assistant / agent. Talks to a model backend and orchestrates around the answers.

Web · '40

The CFML tier. Lucee → Tomcat → JVM: bytecode served as servlets through the Jakarta EE web container.

nginx

The shared front door across all of them — and PostgreSQL is the system of record.

Want a deeper look at the platform?

Talk to our engineering team about how the AkimTechCloud stack fits your workloads.

Contact Sales →