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
| Component | Minimum | Notes |
|---|---|---|
| uxopian-ai | 512 Mi RAM, 0.5 CPU | JVM heap: -Xmx768m -Xms512m |
| uxopian-gateway | 256 Mi RAM | JVM heap: -Xmx256m -Xms256m |
| OpenSearch | 1 Gi RAM (single node) | Version 3.3.2 required |
| Java | 21 | Required for bare JAR deployment only |
For multi-replica deployments, both services use Hazelcast for session state distribution. See Kubernetes deployment for cluster configuration.
OpenSearch
OpenSearch 3.3.2 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:
| Registry | Access | Images |
|---|---|---|
artifactory.arondor.cloud:5001 | Credentials required | uxopian-ai, uxopian-gateway, ARender images |
docker.uxopian.com | Public (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).