How do I send feedback to a different workspace?

Context

In multi-service architectures, each service may be scoped to its own LangSmith workspace via LANGSMITH_WORKSPACE_ID. When one service needs to submit feedback for a trace that lives in a different workspace, the create_feedback() call will succeed without errors but the feedback will never appear -- because all API requests are routed to the workspace the client is configured for.

Answer

To send feedback to a run in a different workspace, create a dedicated Client instance per workspace using an org-scoped API key that has access to all relevant workspaces.

Avoid mutating workspace_id on a shared client instance at runtime -- this can cause race conditions when multiple requests are in-flight and may also interfere with LangGraph's internal tracing client.

1. Obtain an org-level API key with access to all target workspaces.

2. At startup, initialize one Client per workspace:

from langsmith import Client

clients = {
    "workspace_a": Client(api_key=ORG_API_KEY, workspace_id="<WORKSPACE_A_ID>"),
    "workspace_b": Client(api_key=ORG_API_KEY, workspace_id="<WORKSPACE_B_ID>")
    }

3. When submitting feedback, use the client that matches the workspace where the target trace lives:

clients["workspace_b"].create_feedback(
    run_id=run_id,
    key="user_score",
    score=1,
    )

Sources

- [Log traces to a project -- Generic cross-workspace tracing](https://docs.langchain.com/langsmith/log-traces-to-project#generic-cross-workspace-tracing)

- [LangSmith Client API reference -- workspace_id](https://reference.langchain.com/python/langsmith/observability/sdk/client/#langsmith.client.Client.workspace_id)