An AI-powered search engine with a generative UI
TypeScript generative-aigenerative-uinextjsollamareactsearxngshadcn-uitailwindcsstavilytypescriptupstashvercel-ai-sdk
import { LangfuseSpanProcessor } from '@langfuse/otel'
import { LangfuseVercelAiSdkIntegration } from '@langfuse/vercel-ai-sdk'
import { registerOTel } from '@vercel/otel'
import { registerTelemetry } from 'ai'
export type Tracing = {
flush: () => Promise<void>
}
/**
* Bootstrap the same span pipeline the app uses, outside Next.js.
*
* Credentials are required rather than optional: the experiment runner uploads
* item scores through the Langfuse client whether or not spans are exported, so
* a keyless run fails partway instead of degrading to a local-only run.
*/
export function setupTracing(): Tracing {
if (!process.env.LANGFUSE_PUBLIC_KEY || !process.env.LANGFUSE_SECRET_KEY) {
throw new Error(
'LANGFUSE_PUBLIC_KEY and LANGFUSE_SECRET_KEY are required. ' +
'Set them in .env.local (see .env.local.example) or in the CI secrets.'
)
}
const processor = new LangfuseSpanProcessor()
registerOTel({
serviceName: 'morphic-evals',
spanProcessors: [processor],
// The default fetch instrumentation corrupts non-streaming provider
// responses outside Next.js ("Invalid JSON response" on most multi-step
// generations). The AI SDK emits its own spans, so evals do not need it.
instrumentations: []
})
registerTelemetry(new LangfuseVercelAiSdkIntegration())
return { flush: () => processor.forceFlush() }
}