@neatlogs/instrumentation-google-genai

OpenTelemetry instrumentation for the Google GenAI SDK (@google/genai).

Automatically traces generateContent() and generateContentStream() calls with OpenInference-compatible attributes.

Installation

npm install @neatlogs/instrumentation-google-genai
# or
pnpm add @neatlogs/instrumentation-google-genai

Usage

import { NodeTracerProvider } from '@opentelemetry/sdk-trace-node';
import { SimpleSpanProcessor, ConsoleSpanExporter } from '@opentelemetry/sdk-trace-base';
import { GoogleGenAIInstrumentation } from '@neatlogs/instrumentation-google-genai';

const provider = new NodeTracerProvider();
provider.addSpanProcessor(new SimpleSpanProcessor(new ConsoleSpanExporter()));
provider.register();

const instrumentation = new GoogleGenAIInstrumentation();
instrumentation.enable();

// Now all @google/genai calls are automatically traced
import { GoogleGenAI } from '@google/genai';
const ai = new GoogleGenAI({ apiKey: process.env.GOOGLE_API_KEY });
const response = await ai.models.generateContent({
  model: 'gemini-2.0-flash',
  contents: [{ role: 'user', parts: [{ text: 'Hello!' }] }],
});

Captured Attributes

| Attribute | Description | |-----------|-------------| | openinference.span.kind | Always "LLM" | | gen_ai.system | Always "google_genai" | | gen_ai.request.model | Model name | | llm.model_name | Model name | | llm.input_messages.{i}.message.role | Input message role | | llm.input_messages.{i}.message.content | Input message content | | llm.output_messages.{i}.message.role | Output message role | | llm.output_messages.{i}.message.content | Output message content | | llm.token_count.prompt | Prompt token count | | llm.token_count.completion | Completion token count | | llm.token_count.total | Total token count | | gen_ai.request.temperature | Temperature parameter | | gen_ai.request.max_tokens | Max output tokens | | gen_ai.request.top_p | Top-p parameter | | gen_ai.request.top_k | Top-k parameter | | gen_ai.response.finish_reasons | Finish reasons array |