The Langfuse plugin for Genkit provides observability and tracing for your Genkit applications by automatically exporting traces to Langfuse.
npm install genkit-langfuse
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
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
})
],
});
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 */],
});
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';
},
});
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
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'
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
Automatic cost estimation for popular models, or provide custom cost calculation:
calculateCost: (modelName, usage) => {
// Your custom cost calculation logic
return estimatedCost;
}
- 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)
- Flow name and input/output
- Total execution time
- Session context
- Nested span hierarchy
- Complete request lifecycle
- Operation name and type (tools, utilities, etc.)
- Input/output data
- Execution timing
- Parent-child relationships
- Genkit-specific metadata
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
});
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
If traces aren't appearing in Langfuse:
- Check debug logs for HTTP errors
- Verify credentials are correct
- Ensure
forceDevExport: true
in development - Check network connectivity to Langfuse API
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 |
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
The plugin includes full TypeScript support with detailed type definitions for configuration options and metadata structures.
- Firebase - Official Twitter account for Firebase.
- Genkit Discord server - Official Discord server for Genkit.
- Genkit GitHub - Official GitHub repository for Genkit.