LLM Observability: Losing all type hints when using PostHog OpenAI wrapper

[alexeichhorn]

Following this tutorial posthog.com/docs/llm-analytics/installation/openai, we setup everything in our existing quite large codebase.

However, we noticed already on the first example (exactly copied from tutorial):

response = client.responses.create(
    model="gpt-4o-mini",
    input=[
        {"role": "user", "content": "Tell me a fun fact about hedgehogs"}
    ],
    posthog_distinct_id="user_123", # optional
    posthog_trace_id="trace_123", # optional
    posthog_properties={"conversation_id": "abc123", "paid": True}, # optional
    posthog_groups={"company": "company_id_in_your_db"},  # optional 
    posthog_privacy_mode=False # optional
)

Our type-checker (pyright) reports:
error: No overloads for "create" match the provided arguments Argument types: (Literal['gpt-4o-mini'], list[dict[str, str]], Literal['user_123'], Literal['trace_123'], dict[str, Unknown], dict[str, str], Literal[False]) (reportCallIssue

Removing all posthog_-prefixed fields makes everything work nicely again.
To fix this we now chose to use it as kwargs instead:

posthog_args = {
    "posthog_distinct_id": "user_123",
    "posthog_trace_id": "trace_123",
    "posthog_properties": {"conversation_id": "abc123", "paid": True},
    "posthog_groups": {"company": "company_id_in_your_db"},
    "posthog_privacy_mode": False,
}

response = self.sync_openai_client.responses.create(
    model="gpt-4o-mini",
    input=[{"role": "user", "content": "Tell me a fun fact about hedgehogs"}],
    **posthog_args,
)

This no longer gives a type error, but now the response parameter has its type complete erased. It's now a Unknown.
This is our current setup that we have running now. Everything "works", BUT since we have a large codebase, it's just a matter of time something changes from OpenAI library side or maybe our side and something breaks. Usually no problem thanks to type-checks. But now, we won't notice it, since all types are gone.
So, we walking currently on a very dangerous path.

Would appreciate any comment on if this is planned to be improved or fixed. Would be very valuable to us.