NeuralTrust Platform runs on Azure Kubernetes Service using Azure-native primitives — Application Gateway Ingress Controller (AGIC) or NGINX for ingress, Azure Disk for persistent storage, Key Vault for certificates, and Azure DNS for hostnames.Documentation Index
Fetch the complete documentation index at: https://docs.neuraltrust.ai/llms.txt
Use this file to discover all available pages before exploring further.
Pick your path
Hybrid (recommended)
Data Plane + TrustGate + Firewall in your AKS cluster. Control Plane runs on NeuralTrust SaaS. Fastest to first dashboard.
Self-hosted
Full stack including Control Plane API, UI, and Scheduler in your AKS cluster. For sovereignty and air-gapped requirements.
Cluster prerequisites
| Resource | Recommended starting point |
|---|---|
| AKS version | 1.28 or newer |
| CPU pool node SKU | Standard_D8s_v5 or Standard_D8ds_v5 (8 vCPU / 32 GiB) |
| Min CPU nodes | ≥ 4 for hybrid (CPU Firewall), ≥ 5 for self-hosted (CPU Firewall). Subtract one when using GPU Firewall workers. Across 3 AZs for HA. |
| GPU pool (optional) | Standard_NC4as_T4_v3 (4 vCPU / 28 GiB / 1 × T4) — 5 nodes (one per default Firewall worker) |
| VNet | Dedicated VNet with subnets for AKS and Application Gateway (if using AGIC) |
| Storage | Azure Disk CSI driver (default); managed-csi-premium for production |
| Ingress | AGIC, NGINX, or any conformant ingress controller |
| DNS | Azure DNS or any DNS provider |
| Certificates | Key Vault certificate referenced by AGIC, or pre-existing TLS secrets |
Standard_D4s_v5 (4 vCPU / 16 GiB) workers also work but require 7–8 nodes to fit the same workload. See Deployment models › Sizing baseline for the math.
For GPU Firewall workers, add a Standard_NC* or Standard_ND* node pool with the NVIDIA device plugin.
Required cluster setup
Ingress add-on
- AGIC (managed)
- NGINX
Architecture
All workloads run inside your Azure subscription and VNet. Data never leaves your environment.Azure-specific defaults
Whenglobal.platform: "azure":
- Ingress class:
azure-application-gateway(with AGIC annotations). - TLS: AGIC integrates Key Vault certificates via
appgw.ingress.kubernetes.io/appgw-ssl-certificate; NGINX uses standardkubernetes.io/tlssecrets (typically cert-manager-issued). - Storage class:
managed-csi-premiumrecommended for production;managed-csifor cost-sensitive non-prod.
Common configuration
AGIC annotations
appgw-ssl-certificate workflow.
NGINX + cert-manager
Storage class
Internal-only ingress
For private AKS clusters and VNet-internal endpoints, AGIC with a private Application Gateway, or NGINX with an internal LB:GPU node pool for Firewall workers
Region availability
NeuralTrust runs in any Azure commercial region with AKS support. Choose the region closest to your traffic and target LLM endpoints, or one that meets your data-residency obligations (e.g. EU GDPR boundaries). For Azure Government clouds or sovereign-cloud regions, contact [email protected].Backup and data lifecycle
For production, configure backups against the persistent stores:- PostgreSQL: use Azure Database for PostgreSQL Flexible Server with built-in PITR; disable
neuraltrust-control-plane.infrastructure.postgresql.deploy. - ClickHouse: enable
clickhouse.backup.enabled: truewithazblobstorage, or run ClickHouse Cloud externally. - Kafka: use Confluent Cloud or Azure Event Hubs (Kafka surface); set
infrastructure.kafka.deploy: false.
Verification
Common issues
| Symptom | Likely cause | Fix |
|---|---|---|
| Ingress doesn’t get an IP / FQDN | AGIC not enabled or AG subnet undersized | az aks show -g <RG> -n <NAME> --query addonProfiles.ingressApplicationGateway |
PVC stuck Pending | Wrong storage class or quota exhausted | kubectl get storageclass; check subscription quota |
| AGIC cert error | Key Vault cert not synced to AG | Check AGIC logs and Key Vault permissions |
ImagePullBackOff | Missing or wrong gcr-secret | Recreate with the JSON key from NeuralTrust |
Next steps
- Hybrid deployment on AKS — Data Plane only, Control Plane on SaaS
- Self-hosted deployment on AKS — full stack including CP
- Deployment models — hybrid vs self-hosted in depth
- Feature flags reference — local vs external Postgres/Redis/Kafka/CH, image registry, storage, secrets
- Image catalog — what runs where
- Firewall deployment — GPU workers on AKS