aip_sdk.resolve_metric_configs

aip_sdk.resolve_metric_configs(*, workspace_id: str, metric_name: str | None = None, client: APIClient | None = None) -> list[MetricConfigResolution]

Resolve what each agentic metric would use if a run were created in this workspace right now.

Only agentic metrics have a workspace-published default; a non-agentic metric is simply absent, not excluded. For each covered metric, this is the current default plus every named override available to pin instead.

Only overrides entries are valid metric_config_refs targets — default is not a Table B row metric_config_refs can pin (see MetricConfigResolution). Omit the metric from metric_config_refs to use the default; pin one of overrides' pairs for anything else.

One unpaginated response per call; each override's version history is capped (see MetricConfigResolutionOverrideGroup) and carries params + prompt slot names, not bodies. Pass metric_name to narrow it when a workspace holds many.

Parameters

• workspace_id str: The workspace to resolve against.
• metric_name str | None: Only this metric; omit to resolve every agentic metric the workspace has a default for.
• client APIClient | None: Optional pre-configured API client.

Returns

• list[MetricConfigResolution]: list[MetricConfigResolution]: one entry per resolved agentic metric. Empty if
• list[MetricConfigResolution]: metric_name names a metric with no workspace default — including any non-agentic
• list[MetricConfigResolution]: metric, which never has one.

Raises

• NotFoundError: If no workspace has this id.
• ForbiddenError: If the caller is not a member of this workspace.
• AuthError: If no credentials are configured.
• APIError: If the read fails, or the response is malformed.

Example:

for resolution in aip.resolve_metric_configs(workspace_id="ws-abc123"):
    print(resolution.metric_name, resolution.preselected_config_version)