Deployment Options | 4MINDS
Product · Deployment

Runs inside your perimeter.

On your infrastructure. Managed by us. Or in your own cloud account. Three production-grade options — pick the one that fits your architecture.

View pricing →
SELF-MANAGED

On Your Infrastructure

Deploy on your Kubernetes cluster, on-prem or air-gapped. Ghost Weights trains continuously on your data inside your perimeter — with an eval gate before every update. Infrastructure sovereignty and perpetual learning, together.

  • Air-gapped capable
  • Kubernetes-native
  • Zero external API calls
  • Full data sovereignty
  • Classified enclave capable
  • Perpetual learning (Ghost Weights)
FASTEST TO PRODUCTION

Managed by 4MINDS

4MINDS runs and maintains the platform on AWS, Google Cloud, or Azure. Fastest path to production — no Kubernetes team required.

  • SLA-backed uptime
  • Managed upgrades
  • AWS / GCP / Azure
  • No infra team required
  • Available on cloud marketplaces
BYOC

Your Cloud, Our Software

Deploy into your own AWS, Google Cloud, or Azure account. You own the infrastructure. 4MINDS manages the software.

  • Your VPC / VNet
  • You control billing
  • Software-only license
  • IAM integration
  • VPC endpoint support
Deploy-Once vs. Perpetual Learning

Most on-prem AI deployments deploy once. Your model is frozen the day it ships. 4MINDS is different: Ghost Weights trains continuously on your data inside your perimeter — with an eval gate and audit trail before every production update. You get infrastructure sovereignty and a model that gets smarter about your business over time. Red Hat handles the former. 4MINDS handles both.

Already on RHEL / OpenShift?

4MINDS runs on your existing RHEL/OpenShift Kubernetes infrastructure. You keep your Red Hat investment. You add Ghost Weights, Graph RAG, and a native agent platform that shares context with your fine-tuned model.

Infrastructure requirements

Minimum specs for a single-node evaluation deployment. Production deployments scale horizontally across multiple GPU nodes.

Minimum GPU
A100 40GB (single-node POC)
Recommended GPU
4× A100 80GB per inference node
Kubernetes version
1.27+
Container runtime
containerd 1.6+ / CRI-O
Storage
500GB+ NVMe per GPU node (model weights)
Network
25GbE minimum between nodes
OS
Ubuntu 22.04 LTS / RHEL 8+ / Rocky Linux 8+

Validate on your infrastructure

Our deployment team will review your infrastructure, answer architecture questions, and walk through a deployment plan tailored to your environment.

See Ghost Weights →See Graph RAG →See the Eval Gate →
From the Blog
All articles →