Release Notes: uxopian-ai v2026.0.0-ft2

Release Date: February 2026 Version: 2026.0.0-ft2

This release introduces dynamic LLM provider management, a standalone Gateway service, and significant improvements to the admin panel, statistics, and developer experience.

🌟 Highlights

🔧 Dynamic LLM Provider Configuration

LLM provider configurations are now dynamic entities stored in OpenSearch, replacing the previous static YAML-only approach. Providers, models, and their parameters can be created, updated, and deleted at runtime — per tenant — without restarting the service.

Key changes:

New provider configuration entity with global settings and per-model overrides.
Full CRUD via the Admin API (/api/v1/admin/llm/provider-conf).
Per-tenant configuration with merge strategies: OVERWRITE, MERGE, CREATE_IF_MISSING.
AES-GCM encryption for API secrets at rest.
YAML bootstrapping still supported — configurations defined in llm-clients-config.yml are loaded into OpenSearch at startup, then managed dynamically.

See Configuration Files — Dynamic Provider Configuration and LLM Provider Management.

🛡️ Standalone Gateway Service

The BFF Gateway is now a standalone service, deployed independently from the AI service. The runtime architecture remains the same (Gateway authenticates, injects headers, proxies to AI service), but the Gateway can now be scaled and updated independently. This simplifies deployment and allows independent scaling of the security layer.

See Security Model.

📊 Statistics Improvements

The statistics API has been expanded from a single endpoint to 5 dedicated endpoints, each with a configurable time interval parameter:

GET /api/v1/admin/stats/global — Aggregate counters
GET /api/v1/admin/stats/timeseries?interval=DAY — Time-series trends
GET /api/v1/admin/stats/llm-distribution — Model usage breakdown
GET /api/v1/admin/stats/top-prompts-time-saved — ROI ranking
GET /api/v1/admin/stats/feature-adoption — Advanced feature usage rates

Supported intervals: HOUR, DAY, WEEK, MONTH, YEAR.

See Statistics & ROI and REST API Reference.

✨ New Features

🧪 Prompt Tester

The admin panel now includes a Prompt Tester that lets you execute prompts directly from the UI:

Automatically detects Thymeleaf variables in the prompt template.
Provides input fields for each variable (text or image).
Executes the prompt against the configured LLM and displays the result.
Generates the equivalent cURL command for easy reproduction.

See Prompt Management — Prompt Tester.

🖥️ LLM Provider Admin UI

A complete management interface for LLM provider configurations:

Provider List — Table with search, filter, and CRUD actions.
Provider Editor — Form to configure provider identity, global settings, and per-model overrides.
Connection Tester — Test connectivity per model with live status badges.

See LLM Provider Management.

🔑 Fast2 Authentication Provider

New built-in Fast2Provider for the Gateway. It validates JWT tokens issued by Fast2 by fetching the public key from a configurable remote endpoint. Configure it as provider: Fast2Provider in the Gateway route configuration.

📖 Swagger / OpenAPI Improvements

All admin controllers now use consistent Admin - tag prefixes for better organization.
Complete request/response schema documentation on all endpoints.
Swagger UI is publicly accessible (no authentication required) — ideal for API exploration during development.

🏗️ Technical Changes

🤖 ModelProvider Interface Update

The ModelProvider interface has been simplified:

Before: createChatModelInstance(String modelName), getDefaultModelName(), getSupportedModels()
After: createChatModelInstance(LlmModelConf params), createStreamingChatModelInstance(LlmModelConf params)

The getDefaultModelName() and getSupportedModels() methods have been removed — model metadata is now managed via dynamic provider configurations. Custom providers should extend AbstractLlmClient and use params.getModelName(), params.getApiSecret(), etc.

See Adding a New LLM Provider.

⚖️ Parameter Precedence (5 Levels)

The parameter resolution hierarchy has been extended from 3 to 5 levels:

API Call Parameters — Values passed directly in the request.
Prompt Defaults — defaultLlmModel, defaultLlmProvider on the Prompt entity.
Provider Model Config — Per-model settings in LlmModelConf.
Provider Global Config — Shared settings in LlmProviderConf.globalConf.
YAML Global Defaults — llm.default.* in llm-clients-config.yml.

📦 Dependency Upgrades

Dependency	Previous	Current
Spring Boot	3.5.x	3.5.10
LangChain4J	1.x	1.11.0
OpenSearch Client	2.x	3.5.0
Docker Base Image	1.0.x	1.0.4

🎨 Frontend Changes

Improved state management for better performance and responsiveness.
Fixed edge cases in markdown rendering within chat responses.
Improved auto-scroll behavior during streaming responses.

🔄 Migration Notes

From v2026.0.0-ft1-rc2

LLM Configuration: The llm-clients-config.yml format has changed. The previous supported-models lists under each provider section are no longer supported. You must migrate your provider and model definitions to the new llm.provider.globals / llm.provider.tenants structure. These configurations are loaded into OpenSearch at startup and can then be managed dynamically via the Admin API or UI. See Configuration Files — Dynamic Provider Configuration for the new format and a full YAML example.
Gateway Deployment: The Gateway is now deployed as a separate service. Update your Docker compose to use the dedicated Gateway image (uxopian-ai/gateway-service). The configuration format (application.yml with routes) remains the same.
Custom LLM Providers: If you have custom ModelProvider implementations, update them to accept LlmModelConf instead of String in factory methods. Extend AbstractLlmClient for convenience. See the updated guide.
API Secret Encryption: Set the APP_SECURITY_SECRET_KEY environment variable (Base64-encoded AES key) to enable encryption of provider API secrets stored in OpenSearch. If not set, secrets are stored in clear text.

Ready to start? Check out the Quick Start or the full Installation Guide.