Langfuse for Genkit

The Langfuse plugin for Genkit provides observability and tracing for your Genkit applications by automatically exporting traces to Langfuse.

Installation

npm install genkit-langfuse

Configuration

Set up your Langfuse credentials as environment variables:

export LANGFUSE_SECRET_KEY="sk-lf-..."
export LANGFUSE_PUBLIC_KEY="pk-lf-..."
export LANGFUSE_BASE_URL="https://cloud.langfuse.com"  # optional, defaults to cloud

Usage

Plugin Approach (Recommended)

The plugin can be registered with Genkit’s plugin system:

import { genkit } from "genkit"
import { langfuse } from "genkit-langfuse"
 
const ai = genkit({
  plugins: [
    langfuse({
      secretKey: process.env.LANGFUSE_SECRET_KEY!,
      publicKey: process.env.LANGFUSE_PUBLIC_KEY!,
      baseUrl: process.env.LANGFUSE_BASE_URL, // optional
      debug: true, // optional
      forceDevExport: true, // enable in development
    }),
  ],
})

Standalone Telemetry Setup

Alternatively, you can enable telemetry independently (useful for advanced configurations):

import { enableLangfuseTelemetry } from "genkit-langfuse"
import { genkit } from "genkit"
 
// Enable Langfuse telemetry before genkit initialization
await enableLangfuseTelemetry({
  secretKey: process.env.LANGFUSE_SECRET_KEY!,
  publicKey: process.env.LANGFUSE_PUBLIC_KEY!,
  baseUrl: process.env.LANGFUSE_BASE_URL, // optional
  debug: true, // optional
  forceDevExport: true, // enable in development
})
 
const ai = genkit({
  plugins: [
    /* your other plugins */
  ],
})

Advanced Configuration

await enableLangfuseTelemetry({
  secretKey: process.env.LANGFUSE_SECRET_KEY!,
  publicKey: process.env.LANGFUSE_PUBLIC_KEY!,
  baseUrl: "https://your-langfuse-instance.com",
  debug: true,
 
  // Development vs Production settings (following Genkit patterns)
  forceDevExport: true, // Force export in development
  flushAt: 1, // Immediate export in dev (20 in production)
  flushInterval: 1000, // 1s in dev (10s in production)
  exportTimeoutMillis: 10000, // HTTP timeout
  maxQueueSize: 1000, // Max queued spans
 
  // Custom cost calculation
  calculateCost: (modelName, usage) => {
    const rates = {
      "gemini-1.5-flash": { input: 0.075, output: 0.3 },
      "gemini-1.5-pro": { input: 3.5, output: 10.5 },
      "gpt-4": { input: 30, output: 60 },
    }
    const rate = rates[modelName] || { input: 0, output: 0 }
    return (usage.inputTokens * rate.input + usage.outputTokens * rate.output) / 1_000_000
  },
 
  // Custom span filtering
  spanFilter: (span) => {
    // Only export model calls and flows
    return span.spanType === "model" || span.spanType === "flow"
  },
})

Features

Automatic Trace Export

The plugin automatically exports Genkit traces to Langfuse with the following mapping:

• Generations: LLM/model calls are exported as Langfuse generations
• Traces: Root spans and flows are exported as Langfuse traces
• Spans: Intermediate operations are exported as Langfuse spans

Session Tracking

When using Genkit’s chat functionality, session information is automatically included:

const chat = ai.chat({ sessionId: "user-123" })
await chat.send("Hello!")
// Langfuse will receive sessionId: 'user-123'

Rich Metadata

The plugin captures comprehensive metadata including:

• Model name and provider
• Token usage (input/output/total)
• Configuration parameters
• Execution state and timing
• Session and thread information
• Custom metadata from spans

Cost Tracking

Automatic cost estimation for popular models, or provide custom cost calculation:

calculateCost: (modelName, usage) => {
  // Your custom cost calculation logic
  return estimatedCost
}

What Gets Exported

Generations (LLM Calls)

• Model name and provider (e.g., googleai/gemini-1.5-flash)
• Input prompts and configuration
• Output responses and token usage
• Execution timing and performance metrics
• Session and user context
• Cost estimation (if configured)

Traces (Flows/Root Operations)

• Flow name and input/output
• Total execution time
• Session context
• Nested span hierarchy
• Complete request lifecycle

Spans (Intermediate Operations)

• Operation name and type (tools, utilities, etc.)
• Input/output data
• Execution timing
• Parent-child relationships
• Genkit-specific metadata

Development & Debugging

Development Mode

The plugin automatically detects development environment and uses faster export settings:

// Automatically enables immediate export in development
await enableLangfuseTelemetry({
  secretKey: process.env.LANGFUSE_SECRET_KEY!,
  publicKey: process.env.LANGFUSE_PUBLIC_KEY!,
  forceDevExport: true, // Force development behavior
  debug: true, // Enable detailed logging
})

Debug Output

With debug: true, you’ll see detailed logs including:

• Span lifecycle (start/end)
• Export batches and success status
• HTTP communication status
• Langfuse API responses
• Performance metrics

Troubleshooting

If traces aren’t appearing in Langfuse:

1. Check debug logs for HTTP errors
2. Verify credentials are correct
3. Ensure forceDevExport: true in development
4. Check network connectivity to Langfuse API

Configuration Options

Option	Type	Required	Default	Description
secretKey	string	✅	-	Langfuse secret key
publicKey	string	✅	-	Langfuse public key
baseUrl	string	❌	https://cloud.langfuse.com	Langfuse API base URL
debug	boolean	❌	false	Enable detailed logging
forceDevExport	boolean	❌	false	Force export in development
flushAt	number	❌	1 (dev) / 20 (prod)	Batch size for exports
flushInterval	number	❌	1000 (dev) / 10000 (prod)	Export interval in ms
exportTimeoutMillis	number	❌	30000	HTTP request timeout
maxQueueSize	number	❌	1000	Maximum queued spans
calculateCost	function	❌	-	Custom cost calculation
spanFilter	function	❌	-	Filter which spans to export

Environment Variables

• LANGFUSE_SECRET_KEY: Your Langfuse secret key (required)
• LANGFUSE_PUBLIC_KEY: Your Langfuse public key (required)
• LANGFUSE_BASE_URL: Langfuse API base URL (optional, defaults to cloud)
• NODE_ENV: Automatically detected for development settings

TypeScript Support

The plugin includes full TypeScript support with detailed type definitions for configuration options and metadata structures.

Follow

• Firebase - Official Twitter account for Firebase.
• Genkit Discord server - Official Discord server for Genkit.
• Genkit GitHub - Official GitHub repository for Genkit.