feat: add LiteLLM as AI gateway provider

[RheagalFire]

Summary

Adds LiteLLM as a new model provider, giving users access to 100+ LLM providers through a single LiteLLMProvider class. Follows the existing BaseModelProvider pattern exactly.

Motivation

OpenAgents currently supports several providers (OpenAI, Anthropic, Bedrock, Gemini, MiniMax) with dedicated classes, plus a SimpleGenericProvider for OpenAI-compatible APIs. LiteLLM is a lightweight Python SDK that routes requests to the correct provider based on the model string (e.g. anthropic/claude-sonnet-4-5, vertex_ai/gemini-pro, cohere/command-r-plus). This gives users access to providers like Vertex AI, Cohere, AI21, and self-hosted endpoints without needing new provider classes for each one.

Changes

• sdk/src/openagents/lms/providers.py - new LiteLLMProvider class extending BaseModelProvider
• sdk/src/openagents/lms/__init__.py - registered in imports and __all__
• sdk/src/openagents/config/llm_configs.py - added LITELLM enum value, MODEL_CONFIGS entry, and create_model_provider factory case
• pyproject.toml - added litellm>=1.55.0,<1.85 to sdk optional dependencies

Implementation

LiteLLMProvider follows the same pattern as OpenAIProvider:

• chat_completion() calls litellm.acompletion() and returns the standardized {"content": ..., "tool_calls": [...], "usage": {...}} format
• format_tools() uses OpenAI-compatible format (same as OpenAIProvider)
• Tool calling support via tools and tool_choice parameters
• Token usage extraction from the response

Example usage

from openagents.config.llm_configs import create_model_provider
import asyncio

# Use any LiteLLM-supported provider
provider = create_model_provider("litellm", "anthropic/claude-sonnet-4-5")

result = asyncio.run(provider.chat_completion(
    messages=[{"role": "user", "content": "Hello!"}]
))
print(result["content"])

Provider-specific API keys are read from environment variables automatically (e.g. ANTHROPIC_API_KEY, OPENAI_API_KEY). See https://docs.litellm.ai/docs/providers for the full list.

Tests

>>> provider = create_model_provider("litellm", "anthropic/claude-sonnet-4-5")
>>> result = asyncio.run(provider.chat_completion(
...     messages=[{"role": "user", "content": "What is 2+2? Answer with just the number."}]
... ))
>>> result["content"]
'4'
>>> result["usage"]
{'prompt_tokens': 20, 'completion_tokens': 5, 'total_tokens': 25}

Risk / Compatibility

• Additive only. Existing providers untouched.
• litellm>=1.55.0,<1.85 added to the sdk optional dependencies in pyproject.toml.
• Response format matches the existing standardized format used by all other providers.

[vercel]

@RheagalFire is attempting to deploy a commit to the Raphael's projects Team on Vercel.

A member of the Team first needs to authorize it.