AgentBill Vercel AI SDK Integration

OpenTelemetry-based provider wrapper for automatically tracking and billing Vercel AI SDK usage.

Installation

npm install @agentbill/vercel-sdk ai

With specific AI SDK providers:

npm install @ai-sdk/openai @ai-sdk/anthropic

Quick Start

import { AgentBillVercel } from '@agentbill/vercel-sdk';
import { openai } from '@ai-sdk/openai';
import { generateText } from 'ai';

// 1. Initialize AgentBill wrapper
const agentBill = new AgentBillVercel({
  apiKey: 'agb_your_api_key_here',  // Get from AgentBill dashboard
  baseUrl: 'https://bgwyprqxtdreuutzpbgw.supabase.co',
  customerId: 'customer-123',
  debug: true
});

// 2. Wrap your AI SDK provider
const trackedOpenAI = agentBill.wrapProvider(openai);

// 3. Use normally - everything auto-tracked!
const { text } = await generateText({
  model: trackedOpenAI('gpt-4o-mini'),
  prompt: 'What is the capital of France?',
});

console.log(text);

// ✅ Automatically captured:
// - Prompt text (hashed for privacy)
// - Model name (gpt-4o-mini)
// - Provider (openai)
// - Token usage (prompt + completion)
// - Latency (ms)
// - Costs (calculated automatically)

Features

• ✅ Zero-config instrumentation - Just wrap the provider
• ✅ Automatic token tracking - Captures all AI SDK calls
• ✅ Multi-provider support - OpenAI, Anthropic, any AI SDK provider
• ✅ Streaming support - Tracks token-by-token streaming
• ✅ Cost calculation - Auto-calculates costs per model
• ✅ Prompt profitability - Compare costs vs revenue
• ✅ OpenTelemetry compatible - Standard observability

Advanced Usage

Streaming with Tracking

import { streamText } from 'ai';

const result = await streamText({
  model: trackedOpenAI('gpt-4o-mini'),
  prompt: 'Write a poem about code',
});

// Stream to client - all tokens tracked!
for await (const chunk of result.textStream) {
  process.stdout.write(chunk);
}

// Automatically tracked:
// - Total streaming tokens
// - Time to first token
// - Total latency

Track Custom Revenue

// Track revenue for profitability analysis
agentBill.trackRevenue({
  event_name: 'ai_completion',
  revenue: 0.50,  // What you charged the customer
  metadata: { subscription_tier: 'pro' }
});

Use with Multiple Providers

import { anthropic } from '@ai-sdk/anthropic';

const trackedAnthropic = agentBill.wrapProvider(anthropic);

// Mix and match providers - all tracked!
const openaiResult = await generateText({
  model: trackedOpenAI('gpt-4o'),
  prompt: 'Hello from OpenAI'
});

const anthropicResult = await generateText({
  model: trackedAnthropic('claude-3-5-sonnet-20241022'),
  prompt: 'Hello from Anthropic'
});

Next.js API Route Example

// app/api/chat/route.ts
import { AgentBillVercel } from '@agentbill/vercel-sdk';
import { openai } from '@ai-sdk/openai';
import { streamText } from 'ai';

const agentBill = new AgentBillVercel({
  apiKey: process.env.AGENTBILL_API_KEY!,
  baseUrl: process.env.AGENTBILL_BASE_URL!,
  debug: process.env.NODE_ENV === 'development'
});

export async function POST(req: Request) {
  const { messages } = await req.json();
  
  const trackedOpenAI = agentBill.wrapProvider(openai);
  
  // Stream response with automatic tracking
  const result = await streamText({
    model: trackedOpenAI('gpt-4o-mini'),
    messages,
  });
  
  return result.toDataStreamResponse();
}

Configuration

const agentBill = new AgentBillVercel({
  apiKey: 'agb_...',              // Required - get from dashboard
  baseUrl: 'https://...',         // Required - your AgentBill instance
  customerId: 'customer-123',     // Optional - for multi-tenant apps
  accountId: 'account-456',       // Optional - for account-level tracking
  debug: true,                    // Optional - enable debug logging
  batchSize: 10,                  // Optional - batch signals before sending
  flushInterval: 5000             // Optional - flush interval in ms
});

How It Works

The provider wrapper intercepts AI SDK calls:

1. Wrap Provider - Creates a proxy around the AI SDK provider
2. Track Call - Captures model, prompt, start time
3. Execute - Runs the actual AI SDK call
4. Capture Response - Extracts tokens, latency, response
5. Send Signal - Sends data to AgentBill via record-signals API

All data is batched and sent efficiently to minimize overhead.

Supported Models

Auto-cost calculation for:

• OpenAI: GPT-4o, GPT-4o-mini, GPT-4-turbo, GPT-3.5-turbo
• Anthropic: Claude 3.5 Sonnet, Claude 3 Opus, Claude 3 Haiku
• Any AI SDK-compatible provider

TypeScript Support

Full TypeScript support with proper typing:

import type { AgentBillConfig } from '@agentbill/vercel-sdk';

const config: AgentBillConfig = {
  apiKey: 'agb_...',
  baseUrl: 'https://...',
  customerId: 'customer-123'
};

Troubleshooting

Not seeing data in dashboard?

1. Check API key is correct
2. Enable debug: true to see logs
3. Verify baseUrl matches your instance
4. Check network connectivity to AgentBill

TypeScript errors?

Make sure you have the latest version: npm install @agentbill/vercel-sdk@latest

Streaming not working?

The wrapper supports streaming - make sure you're using the latest AI SDK version

License

MIT