Skip to main content

Requirements

Uxopian AI consists of two services: uxopian-ai (the AI backend) and uxopian-gateway (the routing and authentication layer). Both must be deployed and reachable from each other.

Figure: Prerequisite checklist before choosing a deployment model.

Infrastructure​

ComponentMinimumNotes
uxopian-ai512 Mi RAM, 0.5 CPUJVM heap: -Xmx768m -Xms512m
uxopian-gateway256 Mi RAMJVM heap: -Xmx256m -Xms256m
OpenSearch1 Gi RAM (single node)Version 3.6.0 required
Java21 or laterRequired for bare JAR deployment only. Container images ship their own Temurin 25 runtime.

For multi-replica deployments, both services use Hazelcast for session state distribution. See Kubernetes deployment for cluster configuration.

Java runtime​

Both services are compiled for Java 21 bytecode, which is the minimum supported runtime, not a pinned version. Any later JDK runs them.

Deployment modelJava on the hostRuntime actually used
Kubernetes / Docker ComposeNot requiredTemurin 25 (LTS), embedded in the container images
Bare archiveRequired, 21 or laterThe JDK installed on the host

On Java 24 and later, start the JARs with --enable-native-access=ALL-UNNAMED. Without it the JVM emits restricted native access warnings for the dependencies that call native code. The container images already apply this flag. See Bare archive.

OpenSearch​

OpenSearch 3.6.0 is the supported version. uxopian-ai connects to OpenSearch on port 9200 by default. The index is created automatically on first startup.

For production, allocate at least 1 Gi heap to OpenSearch and configure persistent volumes for the data directory.

Image registry access​

Container images are distributed through two registries:

RegistryAccessImages
artifactory.arondor.cloud:5001Credentials requireduxopian-ai, uxopian-gateway, ARender images
docker.uxopian.comPublic (preview channel)uxopian-ai, uxopian-gateway

For Kubernetes deployments, create an image pull secret named regcred pointing to the registry you use. See Kubernetes deployment.

LLM provider​

At least one LLM provider API key is required. Supported providers: OpenAI, Anthropic, Azure OpenAI, Google Gemini, NuExtract. The provider and model are configured at runtime — no rebuild is needed to switch.

Network​

  • uxopian-gateway must be reachable from user browsers on its configured port (default 8085).
  • uxopian-ai must be reachable from uxopian-gateway (internal network, not exposed externally).
  • uxopian-ai must be able to reach OpenSearch and any external service used by plugins (ARender DSB, FlowerDocs core).