first shim wrapper around litellm

[miguelg719]

It's time to start doing a full rewrite. LiteLLM will help a lot
here is a sample test:

class Joke(BaseModel):
    joke: str
    explanation: str
    setup: str
    punchline: str

class Jokes(BaseModel):
    jokes: list[Joke]

async def main():
    # Build a unified configuration object for Stagehand
    config = StagehandConfig(
        env="BROWSERBASE",
        api_key=os.getenv("BROWSERBASE_API_KEY"),
        project_id=os.getenv("BROWSERBASE_PROJECT_ID"),
        model_name="gemini/gemini-2.5-flash-preview-04-17",
        model_client_options={"apiKey": os.getenv("MODEL_API_KEY")},
        # Use verbose=2 for medium-detail logs (1=minimal, 3=debug)
        verbose=3,
    )

    stagehand = Stagehand(
        config=config, 
        server_url=os.getenv("STAGEHAND_SERVER_URL"),
    )
    response = stagehand.llm.create_response(
        messages=[{"role": "user", "content": "Hello, how are you? can you tell me a few jokes?"}],
        model="gemini/gemini-2.0-flash",
        response_format=Jokes,
    )
    print("Received={}".format(response.choices[0].message.content))

[linear]

STG-339 integrate LiteLLM

[filip-michalsky]

Can we please target this to a different branch than main - I was thinking like v2 or python-rewrite or something. Once we are done, we merge to main? what do you think? this way if we need to fix issues in the "stagehand API wrapper" version, we can still do that in parallel?

[filip-michalsky]

I would probably spec out the lib versions here - litellm moves very fast from my experience and its easy to introduce breaking changes.

[filip-michalsky]

you would need to await this in the async mode, right? this call is otherwise blocking the event loop i think

[miguelg719]

yea good catch, gonna remove acompletion until we find a use case for it

[filip-michalsky]

LGTM -

great start, I assume the next step is to wire up the first stagehand primitive here... maybe observe as Ani said?

please merge to a different branch than main though