feat(studio): move agent eval over to evaluator endpoints

plugins/nemo-agents/examples/email-phishing-analyzer/src/nat_email_phishing_analyzer/email-phishing-agent.yml

@@ -28,7 +28,7 @@ llms:
     api_key: not-used
     model_name: ${NEMO_DEFAULT_MODEL}
     temperature: 0.0
-    max_tokens: 512
+    max_tokens: 1024
 
 workflow:
   _type: react_agent

web/packages/common/src/components/DatasetFileSelect/ControlledDatasetFileSelect.tsx

@@ -39,6 +39,7 @@ interface ControlledDatasetFileSelectProps extends UseControllerComponentProps {
   datasetLabel?: string;
   /** Auto-select the first root-level accepted file on fileset selection. */
   autoSelectFirstAcceptable?: boolean;
+  showUpdatedAt?: boolean;
   /**
    * Callback fired when a file is selected. Useful for custom validation or processing.
    * Called with the selected file info, or null when file is cleared.
@@ -83,6 +84,7 @@ export const ControlledDatasetFileSelect: FC<ControlledDatasetFileSelectProps> =
   filesetPurpose,
   datasetLabel,
   autoSelectFirstAcceptable,
+  showUpdatedAt,
 }) => {
   const {
     field: { onChange, value },
@@ -144,6 +146,7 @@ export const ControlledDatasetFileSelect: FC<ControlledDatasetFileSelectProps> =
         filesetPurpose={filesetPurpose}
         datasetLabel={datasetLabel}
         autoSelectFirstAcceptable={autoSelectFirstAcceptable}
+        showUpdatedAt={showUpdatedAt}
       />
     </FormField>
   );

web/packages/common/src/components/DatasetFileSelect/DatasetFileSelect.tsx

@@ -56,6 +56,7 @@ interface DatasetFileSelectProps {
   datasetLabel?: string;
   /** Auto-select the first root-level accepted file on fileset selection. */
   autoSelectFirstAcceptable?: boolean;
+  showUpdatedAt?: boolean;
 }
 
 /**
@@ -87,6 +88,7 @@ export const DatasetFileSelect: FC<DatasetFileSelectProps> = ({
   filesetPurpose,
   datasetLabel,
   autoSelectFirstAcceptable,
+  showUpdatedAt,
 }) => {
   const [isModalOpen, setIsModalOpen] = useState(false);
 
@@ -211,6 +213,7 @@ export const DatasetFileSelect: FC<DatasetFileSelectProps> = ({
             filesetPurpose={filesetPurpose}
             datasetLabel={datasetLabel}
             autoSelectFirstAcceptable={autoSelectFirstAcceptable}
+            showUpdatedAt={showUpdatedAt}
           />
         ) : (
           <DatasetFileSelectButton

web/packages/common/src/components/UploadModal/Context/useUploadModalReducer.ts

@@ -52,6 +52,7 @@ export type UploadModalState = {
   datasetLabel?: string;
   /** Auto-select the first root-level accepted file on fileset selection. */
   autoSelectFirstAcceptable?: boolean;
+  showUpdatedAt?: boolean;
   errors: Record<string, string>;
 };
 
@@ -106,6 +107,7 @@ export const uploadModalReducer = (
         allowMultipleFileSelection: state.allowMultipleFileSelection,
         invalidFileMode: state.invalidFileMode,
         allowNewDataset: state.allowNewDataset,
+        showUpdatedAt: state.showUpdatedAt,
       };
 
     case 'UPDATE_DATASET':

web/packages/common/src/components/UploadModal/DatasetUploader/Select.tsx

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: Apache-2.0
 
 import { getFileExtension } from '@nemo/common/src/components/DatasetFileSelect/utils';
+import { RelativeTime } from '@nemo/common/src/components/RelativeTime';
 import { useUploadModalContext } from '@nemo/common/src/components/UploadModal/Context/useUploadModalContext';
 import { getExistingFileId } from '@nemo/common/src/components/UploadModal/utils';
 import { getEntityReference } from '@nemo/common/src/namedEntity';
@@ -20,14 +21,29 @@ interface Props {
   error?: string;
 }
 
-const filesetToOption = (fileset: FilesetOutput) => ({
+const filesetToOption = (fileset: FilesetOutput, showUpdatedAt?: boolean) => ({
   children: fileset.name ?? '',
   value: getEntityReference(fileset),
+  ...(showUpdatedAt && fileset.updated_at
+    ? {
+        slotEnd: (
+          <Text kind="body/regular/xs" color="secondary">
+            <RelativeTime datetime={fileset.updated_at} />
+          </Text>
+        ),
+      }
+    : {}),
 });
 
 export const DatasetSelect: FC<Props> = ({ project, disabled, error }) => {
   const [state, dispatch] = useUploadModalContext();
-  const { dataset, allowNewDataset, acceptableFileTypes, autoSelectFirstAcceptable } = state;
+  const {
+    dataset,
+    allowNewDataset,
+    acceptableFileTypes,
+    autoSelectFirstAcceptable,
+    showUpdatedAt,
+  } = state;
   const purpose = state.filesetPurpose ?? 'dataset';
   const label = state.datasetLabel ?? 'Dataset';
 
@@ -105,9 +121,9 @@ export const DatasetSelect: FC<Props> = ({ project, disabled, error }) => {
     return (
       filesets
         ?.sort((filesetA, filesetB) => (filesetA?.name || '').localeCompare(filesetB.name || ''))
-        .map(filesetToOption) || []
+        .map((fileset) => filesetToOption(fileset, showUpdatedAt)) || []
     );
-  }, [filesets, isLoading, isError]);
+  }, [filesets, isLoading, isError, showUpdatedAt]);
 
   return (
     <FormField

web/packages/common/src/components/UploadModal/InlineUploadPicker.tsx

@@ -25,6 +25,7 @@ type InlineUploadPickerProps = Pick<
   | 'filesetPurpose'
   | 'datasetLabel'
   | 'autoSelectFirstAcceptable'
+  | 'showUpdatedAt'
 > & {
   /** Called once the picked / uploaded file is committed. */
   onSubmit: (data: SubmitUploadType) => void;
@@ -154,6 +155,7 @@ export const InlineUploadPicker: FC<InlineUploadPickerProps> = ({
   filesetPurpose,
   datasetLabel,
   autoSelectFirstAcceptable,
+  showUpdatedAt,
   onSubmit,
   addButtonText = 'Add file',
   autoCommit = false,
@@ -178,6 +180,7 @@ export const InlineUploadPicker: FC<InlineUploadPickerProps> = ({
       datasetLabel: datasetLabel ?? uploadModalInitialState.datasetLabel,
       autoSelectFirstAcceptable:
         autoSelectFirstAcceptable ?? uploadModalInitialState.autoSelectFirstAcceptable,
+      showUpdatedAt: showUpdatedAt ?? uploadModalInitialState.showUpdatedAt,
     }),
     [
       allowMultipleFileSelection,
@@ -188,6 +191,7 @@ export const InlineUploadPicker: FC<InlineUploadPickerProps> = ({
       filesetPurpose,
       datasetLabel,
       autoSelectFirstAcceptable,
+      showUpdatedAt,
     ]
   );
   return (

web/packages/common/src/components/UploadModal/types.ts

@@ -72,6 +72,8 @@ export interface UploadModalProps {
   datasetLabel?: string;
   /** Auto-select the first root-level accepted file on fileset selection. */
   autoSelectFirstAcceptable?: boolean;
+  /** Show each fileset's updated-at date in the picker options (opt-in). */
+  showUpdatedAt?: boolean;
   attributes?: {
     ModalRoot?: React.ComponentProps<typeof ModalRoot>;
     ModalContent?: React.ComponentProps<typeof ModalContent>;

web/packages/studio/public/sample-agents/calculator/eval-config.json

@@ -0,0 +1,89 @@
+{
+  "tasks": [
+    {
+      "id": "calc-0",
+      "intent": "Compute the product of 3 and 7 and compare it to the current hour",
+      "inputs": {
+        "instruction": "What is the product of 3 and 7, and is it greater than the current hour?"
+      },
+      "reference": {
+        "product": 21
+      }
+    },
+    {
+      "id": "calc-1",
+      "intent": "Compute the sum of 12 and 8 and compare it to the current day of the week",
+      "inputs": {
+        "instruction": "Is the sum of 12 and 8 greater than the current day of the week in number form?"
+      },
+      "reference": {
+        "sum": 20
+      }
+    },
+    {
+      "id": "calc-2",
+      "intent": "Compute the difference between 50 and 35 and compare it to the current minute",
+      "inputs": {
+        "instruction": "What is the difference between 50 and 35, and is it smaller than the current minute of the hour?"
+      },
+      "reference": {
+        "difference": 15
+      }
+    }
+  ],
+  "metric": {
+    "bundle_kind": "metric-bundle",
+    "bundle_format_version": "v1",
+    "metric_type": "llm-judge",
+    "metadata": {
+      "description": null,
+      "labels": {}
+    },
+    "outputs": [
+      {
+        "name": "accuracy",
+        "description": null,
+        "value_json_schema": {
+          "description": "Continuous numeric metric value.",
+          "title": "ContinuousScore",
+          "type": "number"
+        }
+      }
+    ],
+    "secrets": {},
+    "payload": {
+      "kind": "inline",
+      "metric": {
+        "type": "llm-judge",
+        "model": "default/nvidia-nemotron-3-super-120b-a12b",
+        "prompt_template": {
+          "messages": [
+            {
+              "role": "user",
+              "content": "You are scoring a calculator agent that performs arithmetic and datetime comparisons.\n\nThe expected numeric result for this task is: {{ item.reference }}.\n\nAgent response:\n{{ sample.output_text }}\n\nScore whether the agent's response contains the correct numeric result and a valid datetime comparison.\n- accuracy = 1 if the response contains the correct number (e.g. 21 for 3*7) and makes a valid comparison to the current time value.\n- accuracy = 0 if the numeric result is wrong or the comparison is missing or nonsensical.\n- accuracy = 0.5 if the numeric result is correct but the datetime comparison is absent or unclear.\n\nRespond with a JSON object {\"accuracy\": <value>} where accuracy is 0, 0.5, or 1."
+            }
+          ]
+        },
+        "scores": [
+          {
+            "name": "accuracy",
+            "minimum": 0,
+            "maximum": 1
+          }
+        ],
+        "inference": {
+          "max_tokens": 1024,
+          "extra_body": {
+            "nvext": {
+              "max_thinking_tokens": 256
+            }
+          }
+        },
+        "reasoning": {
+          "end_token": "</think>"
+        }
+      }
+    }
+  },
+  "max_concurrent_tasks": 1
+}

web/packages/studio/public/sample-agents/email-phishing-analyzer/agent.yml

@@ -20,7 +20,7 @@ llms:
     api_key: not-used
     model_name: ${NEMO_DEFAULT_MODEL}
     temperature: 0.0
-    max_tokens: 512
+    max_tokens: 1024
 workflow:
   _type: tool_calling_agent
   tool_names: [email_phishing_analyzer]