@conclave-ai/observability-langfuse

Langfuse sink for Conclave AI's efficiency-gate metrics. Forwards every LLM call's cost / tokens / latency / cache-hit to Langfuse for trace inspection.

Self-hosted is the intended deployment target per decision #13. Cloud works identically — pass the cloud base URL.

Install

pnpm add @conclave-ai/observability-langfuse @conclave-ai/core

Usage

import { EfficiencyGate, MetricsRecorder } from "@conclave-ai/core";
import { LangfuseMetricsSink } from "@conclave-ai/observability-langfuse";

const sink = new LangfuseMetricsSink({
  publicKey: process.env.LANGFUSE_PUBLIC_KEY,
  secretKey: process.env.LANGFUSE_SECRET_KEY,
  baseUrl: process.env.LANGFUSE_BASEURL, // self-hosted URL (optional for cloud)
});

const gate = new EfficiencyGate({
  metrics: new MetricsRecorder({ sink }),
});

Call sink.shutdown() at process exit to flush pending events.

Env variables

| Variable | Required | Purpose | |---|---|---| | LANGFUSE_PUBLIC_KEY | ✓ | Public key from your Langfuse project | | LANGFUSE_SECRET_KEY | ✓ | Secret key from your Langfuse project | | LANGFUSE_BASEURL | optional | Self-hosted URL (defaults to Langfuse cloud) |