Installation

npm install @commet/ai-sdk @commet/node

Quick Start

Wrap any AI SDK model with commetAI to automatically track token usage through Commet.

import { Commet } from '@commet/node';
import { commetAI } from '@commet/ai-sdk';
import { openai } from '@ai-sdk/openai';
import { generateText } from 'ai';

const commet = new Commet({ apiKey: process.env.COMMET_API_KEY });

const result = await generateText({
  model: commetAI(openai('gpt-4o'), {
    commet,
    feature: 'ai_chat',
    customerId: 'cus_123', // or an external ID like 'user_123'
  }),
  prompt: 'Hello!',
});

Every generateText and streamText call automatically reports input tokens, output tokens, and cache tokens to Commet.

Streaming

Works the same way with streamText — usage is reported when the stream finishes.

import { streamText } from 'ai';

const result = streamText({
  model: commetAI(openai('gpt-4o'), {
    commet,
    feature: 'ai_chat',
    customerId: 'cus_123',
  }),
  prompt: 'Explain billing models',
});

Options

| Option | Type | Required | Description | |--------|------|----------|-------------| | commet | Commet | Yes | Commet SDK instance | | feature | GeneratedFeatureCode | Yes | Feature code to track usage against | | customerId | string | Yes | Commet customer ID (cus_*) or external ID | | idempotencyKey | string | No | Prevent duplicate tracking for retries | | onTrackingError | (error: Error) => void | No | Custom error handler. Defaults to console.warn |

How It Works

commetAI wraps the model with AI SDK middleware that intercepts generate and stream completions. After each call, it reports to Commet:

• Input tokens (including cache read/write breakdown)
• Output tokens
• Model ID (automatically detected)

Tracking is fire-and-forget — it never blocks or delays your AI responses.

Requirements

• ai >= 6.0.0
• @commet/node >= 1.6.0

Documentation

Visit commet.co/docs for the full guide.

License

MIT