[feat] add litellm

[shanyu-strix]

LiteLLM integration

Adds LiteLLM provider, allowing users to have drop-in compatibility with LiteLLM without any additional configuration.

Some clarifications

Why not wrap this around existing routers etc? If LiteLLM is used internally, not only:

• Anthropic/OpenAI models will be defined under the same api, but parsed differently. Therefore this would require excessive amounts of if else for defining each provider.
• OpenRouter models wold need to be handled independently
• LiteLLM unlocks hundreds more models, i.e huggingface etc.

Example use

def _build_llm(metadata: dict[str, Any] | None = None) -> tuple[Any, bool]:
    import litellm

    from strix.config.config import resolve_llm_config
    from .litellm.chat import ChatLiteLLM    

    model, api_key, api_base = resolve_llm_config()
    if not model:
        raise ValueError("STRIX_LLM environment variable must be set")

    return ChatLiteLLM(
        model=model,
        api_key=api_key,
        api_base=api_base,
        metadata=metadata
    )

---

Summary by cubic

Added a new ChatLiteLLM provider for drop-in use of any litellm-supported chat model. Supports structured outputs, image/text message serialization, usage reporting, and robust error handling.

• New Features

  • Added ChatLiteLLM with provider autodetection; supports model, api_key, api_base, temperature, max_tokens, max_retries, and metadata.
  • Structured output via response_format JSON Schema using SchemaOptimizer for Pydantic models.
  • Serializes user/system/assistant messages, including images and assistant tool_calls.
  • Returns token usage, stop_reason, and optional reasoning content; retries and maps LiteLLM errors to ModelRateLimitError/ModelProviderError.
  • Exported ChatLiteLLM in browser_use.__init__.

• Dependencies

  • Added litellm>=1.82.2.

^{Written for commit 8976e59. Summary will update on new commits.}

[CLAassistant]

All committers have signed the CLA.

[cubic-dev-ai]

No issues found across 5 files

---

Since this is your first cubic review, here's how it works:

• cubic automatically reviews your code and comments on bugs and improvements
• Teach cubic by replying to its comments. cubic learns from your replies and gets better over time
• Add one-off context when rerunning by tagging @cubic-dev-ai with guidance or docs links (including llms.txt)
• Ask questions if you need clarification on any suggestion