Anvia

@anvia/langfuse ​

@anvia/langfuse connects Anvia runs to Langfuse tracing and adds scoring, evaluation reporting, datasets, experiments, prompt retrieval, and configurable PII redaction.

Install ​

bash
pnpm add @anvia/core @anvia/langfuse
ts
import { Agent } from '@anvia/core'
import { LangfuseClient } from '@anvia/langfuse'

const langfuse = new LangfuseClient({
  publicKey: process.env.LANGFUSE_PUBLIC_KEY,
  secretKey: process.env.LANGFUSE_SECRET_KEY,
  baseUrl: process.env.LANGFUSE_BASE_URL,
  environment: 'production',
})
const tracing = langfuse.observer({ captureMode: 'safe' })

const agent = new Agent({
  id: 'support',
  model: model,
  observability: { observers: { tracing } },
})

Call langfuse.flush() before a short-lived worker exits and langfuse.close() during service termination so queued traces and scores have time to reach Langfuse.

Beyond tracing ​

  • Publish Anvia evaluation results with langfuse.evalReporter().
  • Read and update datasets with langfuse.datasetClient().
  • Run an Anvia evaluation suite as a Langfuse experiment with langfuse.runEvalExperiment().
  • Resolve versioned text or chat prompts with langfuse.promptClient().
  • Redact common and application-specific PII before capture.

Capture and reliability ​

Safe capture is the appropriate default. Bound captured payloads, choose shallow or deep redaction deliberately, and decide whether score publishing or a missing trace should fail the active evaluation workflow.

Compatibility ​

The package peers with @anvia/core and includes the Langfuse OpenTelemetry and tracing libraries. Credentials, network policy, application shutdown, and Langfuse project configuration remain caller-owned.

Next steps ​

Built for Anvia.