feat: add LiteLLM as LLM provider

[RheagalFire]

Overview

Adds LiteLLM as a new LLM plugin, enabling access to 100+ providers (OpenAI, Anthropic, Google, Azure, Bedrock, Ollama, etc.) via a single unified SDK. Follows the exact same plugin pattern as OpenAILLM,
DeepSeekLLM, GeminiLLM, etc.

Type of change

• Bug fix
• New feature
• Documentation improvement
• Bounty issue submission
• Other:

Changes

• src/llm/plugins/litellm.py - new LiteLLM plugin class with:
  • LiteLLMConfig extending LLMConfig
  • litellm.acompletion() for async completion
  • drop_params=True for cross-provider kwargs compatibility
  • Full function/tool calling support (same as OpenAILLM)
  • @AvatarLLMState.trigger_thinking() and @LLMHistoryManager.update_history() decorators
• pyproject.toml - added litellm>=1.60.0,<2.0.0 as optional dependency under [project.optional-dependencies]
• tests/llm/test_litellm.py - 12 unit tests (all passing)

Unit tests (12/12 passing):

$ PYTHONPATH=src python -m pytest tests/llm/test_litellm.py -v
test_plugin_file_exists PASSED                                                                                                                                                                                     
test_has_litellm_config_class PASSED
test_has_litellm_class PASSED                                                                                                                                                                                      
test_litellm_class_inherits_llm PASSED
test_has_ask_method PASSED
test_ask_is_async PASSED                                                                                                                                                                                           
test_uses_drop_params_true PASSED
test_uses_litellm_acompletion PASSED                                                                                                                                                                               
test_lazy_imports_litellm PASSED
test_acompletion_called_with_drop_params PASSED
test_acompletion_forwards_api_key PASSED                                                                                                                                                                           
test_acompletion_handles_tool_calls PASSED
12 passed in 0.02s                                                                                                                                                                                                 

Live E2E test (Azure Foundry, Claude):

import asyncio, os, litellm                                                                                                                                                                                        
                
async def main():
    response = await litellm.acompletion(
        model="anthropic/claude-sonnet-4-6",
        messages=[{"role": "user", "content": "What is 2+2? Reply with just the number."}],
        drop_params=True,                                                                                                                                                                                          
        api_key=os.environ["ANTHROPIC_FOUNDRY_API_KEY"],
        api_base=os.environ["ANTHROPIC_FOUNDRY_BASE_URL"],                                                                                                                                                         
    )                                                                                                                                                                                                              
    print(response.choices[0].message.content)
                                                                                                                                                                                                                   
asyncio.run(main())
# Output: "4"

Example usage:

# In your OM1 agent config YAML:
llm:                                                                                                                                                                                                               
  class_name: LiteLLM
  config:                                                                                                                                                                                                          
    model: "anthropic/claude-sonnet-4-20250514"
    api_key: "${ANTHROPIC_API_KEY}"

from llm.plugins.litellm import LiteLLM, LiteLLMConfig                                                                                                                                                             
                
config = LiteLLMConfig(
    model="anthropic/claude-sonnet-4-20250514",
    api_key=os.environ["ANTHROPIC_API_KEY"],
)                                                                                                                                                                                                                  
llm = LiteLLM(config=config)
response = await llm.ask("What should the robot do next?")                                                                                                                                                         

See https://docs.litellm.ai/docs/providers for 100+ supported model strings.

Checklist

• Code complies with style guidelines
• Self-review completed
• Documentation updated (if applicable)
• Local testing completed
• Tests added/updated (if applicable)

Impact

• Additive only, existing plugins (OpenAI, DeepSeek, Gemini, Ollama, etc.) untouched
• litellm added as optional dependency (pip install .[litellm])
• drop_params=True silently drops provider-unsupported kwargs
• Auto-discovered by OM1's find_module_with_class() since the class inherits from LLM

Additional Information

• No proxy server needed, uses litellm Python SDK directly
• LiteLLM model strings use the provider/model format (e.g. anthropic/claude-sonnet-4-20250514, azure/gpt-4o, bedrock/anthropic.claude-3-haiku)
• Provider API keys are read from standard env vars automatically (ANTHROPIC_API_KEY, OPENAI_API_KEY, etc.)

[RheagalFire]

cc @openminddev