Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.fenra.io/llms.txt

Use this file to discover all available pages before exploring further.

This guide covers how to send Google AI usage data to Fenra.

​
Gemini API

import { GoogleGenerativeAI } from '@google/generative-ai';

const genAI = new GoogleGenerativeAI(process.env.GOOGLE_API_KEY);

async function chat(prompt) {
  const model = genAI.getGenerativeModel({ model: 'gemini-1.5-pro' });
  const result = await model.generateContent(prompt);
  const usage = result.response.usageMetadata;

  // Send to Fenra
  await fetch('https://ingest.fenra.io/usage/transactions', {
    method: 'POST',
    headers: {
      'Content-Type': 'application/json',
      'X-Api-Key': process.env.FENRA_API_KEY
    },
    body: JSON.stringify({
      provider: 'gemini',
      model: 'gemini-1.5-pro',
      usage: [{
        type: 'tokens',
        metrics: {
          input_tokens: usage.promptTokenCount,
          output_tokens: usage.candidatesTokenCount,
          total_tokens: usage.totalTokenCount
        }
      }],
      context: {
        billable_customer_id: process.env.BILLABLE_CUSTOMER_ID
      }
    })
  });

  return result.response;
}

​
Context Caching

When using context caching, include cached tokens: 
usage: [{
  type: 'tokens',
  metrics: {
    input_tokens: usage.promptTokenCount,
    output_tokens: usage.candidatesTokenCount,
    total_tokens: usage.totalTokenCount,
    cached_tokens: usage.cachedContentTokenCount || 0
  }
}]

​
Multimodal Inputs

Gemini handles images, video, and audio natively. Token counts include all modalities, so no separate tracking is needed.

​
Thinking Tokens

For Gemini 2.5 models with thinking enabled, include thinking tokens: 
usage: [{
  type: 'tokens',
  metrics: {
    input_tokens: usage.promptTokenCount,
    output_tokens: usage.candidatesTokenCount,
    total_tokens: usage.totalTokenCount,
    thinking_tokens: usage.thoughtsTokenCount || 0
  }
}]
When using Google Search grounding, track the tool invocations: 
const result = await model.generateContent({
  contents: prompt,
  tools: [{ functionDeclarations: [searchTool] }]
});

// Check if grounding was used
const groundingMetadata = result.response.groundingMetadata;
const searchQueries = groundingMetadata?.webSearchQueries || [];

await fetch('https://ingest.fenra.io/usage/transactions', {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json',
    'X-Api-Key': process.env.FENRA_API_KEY
  },
  body: JSON.stringify({
    provider: 'gemini',
    model: 'gemini-1.5-pro',
    usage: [
      {
        type: 'tokens',
        metrics: {
          input_tokens: usage.promptTokenCount,
          output_tokens: usage.candidatesTokenCount,
          total_tokens: usage.totalTokenCount
        }
      },
      {
        type: 'requests',
        metrics: {
          count: searchQueries.length,
          request_type: 'google_grounding'
        }
      }
    ],
    context: {
      billable_customer_id: process.env.BILLABLE_CUSTOMER_ID
    }
  })
});

​
Supported Models

Fenra supports all Google Gemini models. Common models include:
ModelContext Window
gemini-2.0-flash1M tokens
gemini-1.5-pro2M tokens
gemini-1.5-flash1M tokens
gemini-1.5-flash-8b1M tokens

​
Next Steps