LLM Client Factory

[itomek]

Summary

Refactors LLM client architecture to use provider-specific implementations instead of a monolithic LLMClient class, improving maintainability and extensibility.

Changes

Architecture

• New base interface: LLMClient (ABC) with provider-specific implementations
• Provider pattern: LemonadeProvider, OpenAIProvider, ClaudeProvider
• Factory function: create_client("lemonade") for easy instantiation
• Removed: Monolithic src/gaia/llm/llm_client.py

Key Features

• Model auto-loading before requests in LemonadeClient
• Default temperature of 0.1 for deterministic responses
• Better separation of concerns between providers
• Simplified import: from gaia.llm import LLMClient, create_client

Migration

Existing code using LLMClient should update imports:

# Old
from gaia.llm.llm_client import LLMClient

# New
from gaia.llm import LLMClient, create_client
client = create_client("lemonade")

[Copilot]

Pull request overview

This PR refactors the LLM client architecture from a monolithic LLMClient class to a provider-specific pattern with a factory function for instantiation, improving maintainability and extensibility.

Key Changes

• Introduces an abstract base class LLMClient with provider-specific implementations (LemonadeProvider, OpenAIProvider, ClaudeProvider)
• Adds a factory function create_client() that auto-detects providers from flags or explicit provider names
• Implements proactive model loading in LemonadeClient to prevent 404 errors
• Updates all imports across the codebase from gaia.llm.llm_client to gaia.llm

Reviewed changes

Copilot reviewed 26 out of 26 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
util/lint.py	Updates import check from gaia.llm.llm_client to gaia.llm
tests/unit/test_llm_client_factory.py	Adds comprehensive tests for factory pattern and provider capabilities
tests/unit/test_llm.py	Removes obsolete base URL normalization tests
tests/unit/test_lemonade_model_loading.py	Adds tests for proactive model loading functionality
tests/test_sdk.py	Updates LLMClient imports to new package structure
src/gaia/mcp/mcp_bridge.py	Updates to use create_client() factory
src/gaia/llm/vlm_client.py	Copyright year update only
src/gaia/llm/providers/openai_provider.py	Implements OpenAI provider with chat, generate, and embed support
src/gaia/llm/providers/lemonade.py	Implements Lemonade provider with full method support including vision
src/gaia/llm/providers/claude.py	Implements Claude provider with vision support
src/gaia/llm/providers/init.py	Exports provider classes
src/gaia/llm/llm_client.py	Removes monolithic LLMClient (723 lines deleted)
src/gaia/llm/lemonade_client.py	Adds base_url parameter, _ensure_model_loaded() method, and prompt parameter to load_model()
src/gaia/llm/factory.py	Implements factory function with auto-detection logic
src/gaia/llm/exceptions.py	Adds NotSupportedError exception
src/gaia/llm/base_client.py	Defines abstract base class with default NotSupportedError implementations
src/gaia/llm/init.py	Exports factory, base client, and exception
src/gaia/cli.py	Updates to use factory with explicit "lemonade" provider
src/gaia/chat/sdk.py	Updates to use factory with auto-detection flags
src/gaia/audio/audio_client.py	Updates to use factory
src/gaia/apps/llm/app.py	Updates to use factory
src/gaia/agents/routing/agent.py	Updates to use factory
src/gaia/agents/blender/tests/test_agent_simple.py	Updates imports to use new base_client module
src/gaia/agents/blender/tests/test_agent.py	Updates imports to use base_client
src/gaia/agents/blender/agent_simple.py	Updates to use factory with system_prompt support
setup.py	Adds gaia.llm.providers package

Comments suppressed due to low confidence (1)

tests/test_sdk.py:430

• The test expects chat_completions, get_available_models, and estimate_tokens methods on LLMClient, but the new base_client.py ABC only defines generate, chat, embed, vision, get_performance_stats, load_model, and unload_model. This test will fail because these methods don't exist in the new interface. The test should be updated to match the new API or these methods should be added to the base client if they're still needed.

        # Check methods exist
        assert hasattr(LLMClient, "generate")
        assert hasattr(LLMClient, "chat_completions")
        assert hasattr(LLMClient, "get_available_models")
        assert hasattr(LLMClient, "estimate_tokens")

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[kovtcharov-amd]

Make sure to test all agents and provide a summary table here once complete. Approved conditional on documentation + testing.

[itomek]

Make sure to test all agents and provide a summary table here once complete. Approved conditional on documentation + testing.

Testing

This has been tested against the following GAIA User Guides:

• Hardware Advisor: ✅ Pass
• Chat SDK: ✅ Pass
• Code Agent: ✅ Pass
• Blender Agent: ✅ Pass
• Eval Framework Reference documentations: ✅ Pass