Conversation API alpha2 ignores per-request metadata (model/key/cacheTTL)

[bibryam]

Expected Behavior

Metadata should override any of the matching metadata fields from the component file

Actual Behavior

When calling converse_alpha2, any metadata you pass (e.g., model/key/cacheTTL) is ignored. The runtime still uses whatever is defined in the conversation component YAML; the request-level metadata never overrides the component settings.

Steps to Reproduce the Problem

Here is the modified quickstart I tried

from dapr.clients import DaprClient
from dapr.clients.grpc.conversation import (
    ConversationInputAlpha2,
    ConversationMessage,
    ConversationMessageContent,
    ConversationMessageOfUser,
)

with DaprClient() as d:
    text_input = "What is dapr in 5 words?"
    provider_component = "openai"

    inputs = [
        ConversationInputAlpha2(
            messages=[
                ConversationMessage(
                    of_user=ConversationMessageOfUser(
                        content=[ConversationMessageContent(text=text_input)]
                    )
                )
            ],
            scrub_pii=True,
        ),
    ]

    metadata = {
        "model": "gpt-4o-mini",  # example: only works if the openai component honors this
        "cacheTTL": "10m",       # example: only if the component supports caching/ttl overrides
        "trace-id": "abc123",    # arbitrary key; ignored unless component uses it
    }

    print(f"Input sent: {text_input}")

    response = d.converse_alpha2(
        name=provider_component,
        inputs=inputs,
        temperature=0.7,
        context_id="chat-123",
        metadata=metadata,  # ← per-request metadata
    )

    for output in response.outputs:
        print(f"Output response: {output.choices[0].message.content}")

Release Note

RELEASE NOTE: