Uxopian AI overview

Uxopian AI is a framework for embedding AI assistants inside legacy enterprise applications. It is designed for content management and document-centric systems. It does not replace the host application. It adds a conversational interface, connects to multiple LLM providers, and exposes tools the LLM can call to operate on enterprise data.

Who this documentation is for

Audience	Entry point
System administrator deploying the stack	Quickstart with Docker Compose
Integration developer embedding the chat UI	Embed in a web application
Integration developer embedding the context-aware panel	Embed Quick Prompt
Solution architect evaluating the system	System architecture
ARender integrator	Integrate with ARender
Alfresco integrator	Integrate with Alfresco
FlowerDocs integrator	Integrate with FlowerDocs
Prompt or assistant developer	Prompts and templating
Gateway / route configuration	Configure gateway routes

System components

Figure: High-level component topology.

uxopian-ai

The primary application. Java 21, Spring Boot (reactive WebFlux stack). Handles all business logic: conversations, requests, prompt rendering, tool execution, LLM calls, and the REST/WebSocket API.

uxopian-gateway

Spring Cloud Gateway acting as a reverse proxy. The only public entry point. Authenticates every request using a configurable AuthProvider and forwards identity headers to uxopian-ai. uxopian-ai must never be exposed directly to browsers.

OpenSearch

Primary persistence store. Stores conversations, requests, prompts, LLM provider configurations, and usage metrics. All data is tenant-scoped.

Hazelcast

Distributed in-memory cache used by the gateway for session token caching. Configured via hazelcast.yml.

Supported LLM providers

Nine providers are supported out of the box:

Provider	Key identifier
OpenAI	openai
Anthropic	anthropic
Azure OpenAI	azure-openai
AWS Bedrock	bedrock
Google Gemini	gemini
Mistral AI	mistral-ai
HuggingFace	huggingface
Ollama	ollama
NuExtract	nu-extract

See LLM providers for configuration details.

Integration paths

Three integration paths are available:

• Generic web application: load the JavaScript and CSS bundles from the gateway's /api/web-components/chat/script and /api/web-components/chat/style endpoints, call window.createChat(). See Embed in a web application.
• Quick Prompt context-aware panel: embed the Quick Prompt panel, which follows the user's navigation and offers context-relevant prompts. See Embed Quick Prompt.
• ARender document viewer: adds an AI menu to the ARender top panel. Documents are accessed via the ARender DSB API. See Integrate with ARender.
• FlowerDocs ECM: embeds the chat panel via FlowerDocs scope files. Uses FlowerDocsProvider in the gateway. See Integrate with FlowerDocs.

Extension mechanisms

• Custom tool plugins: write a @ToolService class, package as a shaded JAR, drop in plugins/. The LLM can then call those methods as tools.
• Custom ServiceHelpers: write a @HelperService class, expose it as a named expression object in Thymeleaf prompt templates.
• Custom auth providers: implement the AuthProvider interface in the gateway to support any identity system.
• Prompt customization: define global prompts in prompts.yml, or manage them live via the Admin API.

Key concepts

Concept	Description
Tenant	Primary isolation unit. All data is scoped to a tenant ID.
Conversation	A chat session. Contains a sequence of Requests.
Request	A single LLM round-trip: inputs, rendered prompt, response, token usage.
Prompt	A named Thymeleaf template defining a role and content, versioned with a draft → publish lifecycle.
Script	Admin-managed front-end JavaScript, security-scanned and served to authenticated users.
Plugin	A shaded JAR in plugins/ loaded at startup by IntegrationLoader.

Next steps

1. Follow Quickstart with Docker Compose to run a local stack.
2. Read System architecture for the full component diagram and request flow.
3. Read Authentication and gateway before any deployment.