Get Artifact In Experiment
Get one normalized artifact descriptor by name from an experiment.
Use this tool when the caller already knows artifact_name and needs its
typed descriptor, such as a scalar value, timeseries, or binary metadata.
The returned shape reflects MCP read-side normalization rather than a
round-trip of the original write-time artifact kind.
When to Use:
- Call this when the caller already knows the artifact key and needs the normalized MCP descriptor.
- Call this before downloading binary data, so the workflow can inspect
mime_typeorcontent_urlfirst.
Constraints:
- This endpoint reads one entry from the experiment metrics map and does not reconstruct the original write-time artifact contract.
- When the descriptor exposes
content_url, raw bytes still require a separate content request.
Notes:
- This endpoint does not stream raw bytes.
- Use
get_artifact_content_in_experimentwhen the artifact should be served as binary content.
Args: experiment_id: Parent experiment identifier. artifact_name: Artifact key within experiment metrics. user: Authenticated user resolved from the API key. db: Active database session.
Returns: One normalized artifact descriptor.
Raises:
HTTPException: Raised with 404 if the experiment is unavailable to
the caller or the artifact key is absent.
Authorizations
Path Parameters
Query Parameters
Response
Successful Response
Represent the normalized artifact shape returned by MCP read endpoints.
Attributes:
name: Artifact key in experiment metrics.
kind: Normalized read-side artifact category, typically value,
timeseries, or a base64-derived media family such as image.
value: Scalar or JSON-like artifact value.
mime_type: MIME type for binary or base64-backed artifacts.
content_url: Relative URL used to fetch raw bytes when needed.
auth_hint: Header hint used by clients when dereferencing
content_url.
x: X-axis metadata for timeseries payloads.
y: Y-axis metadata for timeseries payloads.
points: Ordered point list for timeseries payloads.