stella-ops.org/git.stella-ops.org
1
0
Fork 0
Code Issues Pull Requests Actions 2 Releases Wiki Activity
Files
e9aeadc04084353889b1c9dbf4487f5316f4421e
git.stella-ops.org/docs/features/checked/advisoryai/llm-provider-plugin-architecture.md
master e9aeadc040 save checkpoint
2026-02-14 09:11:48 +02:00

3.5 KiB
Raw Blame History

LLM Provider Plugin Architecture (Multi-Provider Inference)

Module

AdvisoryAI

Status

IMPLEMENTED

Description

Pluggable LLM provider architecture with ILlmProvider interface supporting OpenAI, Claude, Gemini, llama.cpp (LlamaServer), and Ollama backends. Includes LlmProviderFactory for runtime selection and configuration validation. Enables sovereign/offline inference by switching to local providers.

Implementation Details

  • Modules: src/AdvisoryAi/StellaOps.AdvisoryAI/Inference/LlmProviders/, src/AdvisoryAi/StellaOps.AdvisoryAI/Chat/Inference/, src/AdvisoryAi/StellaOps.AdvisoryAI.Plugin.Unified/
  • Key Classes:
    • LlmProviderFactory (src/AdvisoryAi/StellaOps.AdvisoryAI/Inference/LlmProviders/LlmProviderFactory.cs) - factory for runtime LLM provider selection
    • OpenAiLlmProvider (src/AdvisoryAi/StellaOps.AdvisoryAI/Inference/LlmProviders/OpenAiLlmProvider.cs) - OpenAI API provider
    • ClaudeLlmProvider (src/AdvisoryAi/StellaOps.AdvisoryAI/Inference/LlmProviders/ClaudeLlmProvider.cs) - Anthropic Claude API provider
    • GeminiLlmProvider (src/AdvisoryAi/StellaOps.AdvisoryAI/Inference/LlmProviders/GeminiLlmProvider.cs) - Google Gemini API provider
    • LlamaServerLlmProvider (src/AdvisoryAi/StellaOps.AdvisoryAI/Inference/LlmProviders/LlamaServerLlmProvider.cs) - local llama.cpp server provider
    • OllamaLlmProvider (src/AdvisoryAi/StellaOps.AdvisoryAI/Inference/LlmProviders/OllamaLlmProvider.cs) - Ollama local inference provider
    • LlmProviderOptions (src/AdvisoryAi/StellaOps.AdvisoryAI/Inference/LlmProviders/LlmProviderOptions.cs) - provider configuration and validation
    • ClaudeInferenceClient (src/AdvisoryAi/StellaOps.AdvisoryAI/Chat/Inference/ClaudeInferenceClient.cs) - Claude-specific chat inference client
    • OpenAIInferenceClient (src/AdvisoryAi/StellaOps.AdvisoryAI/Chat/Inference/OpenAIInferenceClient.cs) - OpenAI-specific chat inference client
    • OllamaInferenceClient (src/AdvisoryAi/StellaOps.AdvisoryAI/Chat/Inference/OllamaInferenceClient.cs) - Ollama-specific chat inference client
    • LocalInferenceClient (src/AdvisoryAi/StellaOps.AdvisoryAI/Chat/Inference/LocalInferenceClient.cs) - local model inference client
    • LlmPluginAdapter (src/AdvisoryAi/StellaOps.AdvisoryAI.Plugin.Unified/LlmPluginAdapter.cs) - unified plugin adapter for LLM providers
    • LlmPluginAdapterFactory (src/AdvisoryAi/StellaOps.AdvisoryAI.Plugin.Unified/LlmPluginAdapterFactory.cs) - factory for creating LLM plugin adapters
    • SystemPromptLoader (src/AdvisoryAi/StellaOps.AdvisoryAI/Chat/Inference/SystemPromptLoader.cs) - loads system prompts for inference clients
  • Interfaces: ILlmProvider, ILlmProviderPlugin, IAdvisoryChatInferenceClient
  • Source: SPRINT_20251226_019_AI_offline_inference.md

E2E Test Plan

  • Configure LlmProviderFactory with multiple providers and verify runtime selection based on configuration
  • Verify OpenAiLlmProvider sends requests to OpenAI API with correct authentication and model parameters
  • Verify ClaudeLlmProvider sends requests to Claude API with correct authentication
  • Verify OllamaLlmProvider connects to local Ollama instance and performs inference
  • Verify LlamaServerLlmProvider connects to local llama.cpp server endpoint
  • Verify LlmProviderOptions validation rejects invalid configurations (missing API keys, invalid endpoints)
  • Verify LlmPluginAdapter provides health checks for configured LLM providers
  • Verify provider failover: when primary provider is unavailable, factory falls back to secondary