How to Track AI Costs by Project in Under 10 Minutes

You just got a $4,000 OpenAI invoice. Which project caused it? Which feature? Which engineer's code? Without attribution, AI spend is a black box — you know the total but not the breakdown.

Provider dashboards show aggregate spend but can't tell you which internal project or feature generated a specific request. That gap is what cost attribution solves.

Install the Model Meter SDK, then add two lines to your app:

import { init } from '@model-meter/sdk';

init({
  apiKey: process.env.MODEL_METER_API_KEY,
  endpoint: process.env.MODEL_METER_ENDPOINT,
  projectId: 'my-app',
  environment: 'production',
});

After init(), every generateText() and streamText() call is captured automatically via OpenTelemetry — no per-call changes needed.

Use the @model-meter/sdk wrapper instead of importing from 'ai' to add an operation label to each call:

import { generateText } from '@model-meter/sdk';

const result = await generateText({
  model: openai('gpt-4o'),
  prompt: 'Summarize this ticket',
  modelMeter: { operation: 'summarize-ticket' },
});

Operations appear as a breakdown dimension in the dashboard — alongside project, provider, model, and environment.

After your first requests, the dashboard shows spend by project, provider, model, and operation — updated in near real-time, not end-of-month. You'll see which feature is spending what, which model is most cost-efficient for each task, and where the anomalies are.

Model Meter on GitHub → · Read the getting started guide →

Model Meter doesn't sit in the critical path of your AI calls. The SDK instruments existing calls via OpenTelemetry and sends telemetry asynchronously. Your AI requests go directly to the provider, so tracking adds no latency and introduces no new failure point. Prompts and completions are never collected.