ADFA-3017 | Add OpenAI-compatible AI backend

[jatezzz]

Description

Adds the new ai-agent-openai plugin as an OpenAI-compatible inference backend for ai-core.

The implementation supports OpenAI and compatible servers such as Ollama, LM Studio, OpenRouter, and llama-server through a configurable base URL. It includes streaming inference, configurable model selection, secure API key storage, connection verification, server presets, error handling, and reasoning-model request compatibility.

The repository documentation is also updated to include the new backend and its configuration, installation, security, and usage details.

Details

• Registers the openai inference backend with ai-core.
• Adds OpenAI-compatible chat completion and SSE streaming support.
• Supports configurable OpenAI-compatible server URLs.
• Adds server presets for OpenAI, Ollama, LM Studio, OpenRouter, and llama.cpp.
• Adds secure API key storage using Android Keystore and AES/GCM.
• Handles API key requirements based on the configured server.
• Adds connection testing and model catalog retrieval.
• Supports editable model selection and remembered models.
• Handles reasoning-model parameter differences and unsupported-parameter retries.
• Adds URL validation and controlled HTTP support for local/private servers.
• Adds backend, settings, error handling, logging, security, and prompt layers.
• Adds plugin documentation and configuration resources.
• Adds JVM unit tests covering request building, model filtering, request tuning, SSE parsing, errors, URL policies, model selection, and connection state.

🎦 Demo in ticket

Ticket

ADFA-3017

Observation

The backend intentionally uses direct HTTP communication instead of an OpenAI SDK to avoid dependency and classloader conflicts with CodeOnTheGo.

Native function calling is not enabled; the backend relies on ai-core's provider-agnostic agent flow. API keys remain optional for compatible local/private servers and are required only when appropriate for the configured endpoint.

[claude]

Claude Code Review

This repository is configured for manual code reviews. Comment @claude review for a one-time review, or @claude review always to subscribe this PR to a review on every future push.

_{Tip: disable this comment in your organization's Code Review settings.}