🪢 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
import { JSONPath } from "jsonpath-plus";
import { MediaReferenceStringSchema } from "./IORepresentation/chatML/types";
/**
* Scans a JSON value for `@@@langfuseMedia:...@@@` reference strings and
* returns each occurrence with the JSONPath of the containing string.
* References must be the entire string value; references embedded in
* surrounding text and malformed references are ignored.
*/
export function findMediaReferences(value: unknown) {
// Dataset item fields are arbitrary JSON, so the root may itself be a reference
if (typeof value === "string") {
const reference = parseReference(value, "$");
return reference ? [reference] : [];
}
if (typeof value !== "object" || value === null) return [];
// The @string() type selector needs no filter expression, so script
// evaluation can stay disabled. MediaReferenceStringSchema decides which
// strings are references.
const nodes = JSONPath({
path: "$..*@string()",
json: value,
resultType: "all",
eval: false,
}) as { value: string; path: string }[];
return nodes
.map((node) => parseReference(node.value, node.path))
.filter((reference) => reference !== null);
}
function parseReference(value: string, jsonPath: string) {
const parsed = MediaReferenceStringSchema.safeParse(value);
return parsed.success ? { ...parsed.data, jsonPath } : null;
}