🪢 Open source agent evals & observability: Trace, evaluate, and improve LLM applications with one open platform.
TypeScript analyticsautogenevaluationlangchainlarge-language-modelsllama-indexllmllm-evaluationllm-observabilityllmopsmonitoringobservabilityopen-sourceopenaiplaygroundprompt-engineeringprompt-managementself-hostedycombinator
/* eslint-disable @typescript-eslint/no-deprecated -- Legacy UI usage retained during migration to design-system components. */
import { useRef, useState } from "react";
import { useStore } from "zustand";
import { parseJsonPrioritised, type BatchActionQuery } from "@langfuse/shared";
import {
Dialog,
DialogContent,
DialogHeader,
DialogTitle,
DialogDescription,
DialogBody,
} from "@/src/components/ui/dialog";
import { api, sendAsPostOption } from "@/src/utils/api";
import { showErrorToast } from "@/src/features/notifications/showErrorToast";
import { classifyTrpcToastError } from "@/src/utils/trpcErrorClassification";
import { DatasetCreateStep } from "./DatasetCreateStep";
import { DatasetMappingEditor } from "./DatasetMappingEditor";
import { StatusStep } from "./StatusStep";
import { createDatasetMappingStore } from "./datasetMappingStore";
import { submitDatasetBatch } from "./submitDatasetBatch";
import type { ObservationPreviewData } from "./types";
type AddObservationsToDatasetDialogProps = {
projectId: string;
onClose: () => void;
onSuccess: () => void;
selectedObservationIds: string[];
query: BatchActionQuery;
selectAll: boolean;
totalCount: number;
isV4: boolean;
exampleObservation: { id: string; traceId: string; startTime?: Date };
};
function normalizeValue(value: unknown) {
if (typeof value !== "string") return value;
const parsed = parseJsonPrioritised(value);
return parsed === undefined ? value : parsed;
}
export function AddObservationsToDatasetDialog(
props: AddObservationsToDatasetDialogProps,
) {
const [source] = useState(() => ({
selectedObservationIds: [...props.selectedObservationIds],
query: props.query,
selectAll: props.selectAll,
totalCount: props.totalCount,
exampleObservation: props.exampleObservation,
isV4: props.isV4,
}));
const [store] = useState(createDatasetMappingStore);
const screen = useStore(store, (state) => state.screen);
const submission = useStore(store, (state) => state.submission);
const createPending = useRef(false);
const count = source.selectAll
? source.totalCount
: source.selectedObservationIds.length;
const example = source.exampleObservation;
const datasets = api.datasets.allDatasetMeta.useQuery({
projectId: props.projectId,
});
const observationQuery = api.observations.byId.useQuery(
{
projectId: props.projectId,
observationId: example.id,
traceId: example.traceId,
startTime: example.startTime,
},
{ enabled: !source.isV4 && Boolean(example.id && example.traceId) },
);
const eventQuery = api.events.batchIO.useQuery(
{
projectId: props.projectId,
observations: [{ id: example.id, traceId: example.traceId }],
minStartTime: example.startTime as Date,
maxStartTime: example.startTime as Date,
truncated: false,
},
{
...sendAsPostOption,
enabled:
source.isV4 &&
Boolean(example.id && example.traceId && example.startTime),
},
);
const raw = source.isV4 ? eventQuery.data?.[0] : observationQuery.data;
const observation: ObservationPreviewData | null = raw
? {
id: example.id,
input: normalizeValue(raw.input),
output: normalizeValue(raw.output),
metadata: normalizeValue(raw.metadata),
}
: null;
const previewQuery = source.isV4 ? eventQuery : observationQuery;
const mutation = api.batchAction.addToDataset.create.useMutation({
onError: (error) =>
showErrorToast(
"Failed to schedule action",
error.message,
classifyTrpcToastError(error, "dataset.add_observations"),
),
});
return (
<Dialog
open
onOpenChange={(open) => {
if (
!open &&
store.getState().submission.status !== "pending" &&
!createPending.current
)
props.onClose();
}}
>
<DialogContent size="xl">
<DialogHeader>
<DialogTitle>
{screen === "create"
? "Create dataset"
: `Add ${count} observation${count === 1 ? "" : "s"} to dataset`}
</DialogTitle>
<DialogDescription>
{screen === "create"
? "Create a dataset, then review the values to add."
: "Choose a dataset and review how each observation becomes a dataset item."}
</DialogDescription>
</DialogHeader>
{submission.status === "scheduled" && (
<DialogBody>
<StatusStep
projectId={props.projectId}
batchActionId={submission.batchActionId}
dataset={submission.dataset}
expectedCount={count}
onClose={props.onClose}
/>
</DialogBody>
)}
{submission.status !== "scheduled" && screen === "create" && (
<DatasetCreateStep
projectId={props.projectId}
onDatasetCreated={(dataset) => {
store.getState().actions.datasetCreated(dataset);
}}
onSubmittingChange={(pending) => {
createPending.current = pending;
}}
onCancel={() => store.getState().actions.setScreen("compose")}
/>
)}
{submission.status !== "scheduled" && screen === "compose" && (
<DatasetMappingEditor
store={store}
datasets={datasets.data ?? []}
loading={datasets.isLoading || previewQuery.isLoading}
unavailable={
datasets.isError ||
previewQuery.isError ||
(!previewQuery.isLoading && !observation)
}
onRetry={() => {
datasets.refetch();
previewQuery.refetch();
}}
observation={observation}
count={count}
onClose={props.onClose}
onSubmit={(dataset) =>
submitDatasetBatch({
projectId: props.projectId,
store,
dataset,
observation,
source,
submit: mutation.mutateAsync,
onSuccess: props.onSuccess,
})
}
/>
)}
</DialogContent>
</Dialog>
);
}