@agent-qa/cost-registry

Unified cost registry for LLM pricing and token counting across providers.

Installation

npm install @agent-qa/cost-registry
# or
pnpm add @agent-qa/cost-registry

Features

• Calculate LLM costs from token usage (Anthropic, OpenAI)
• Provider-specific pricing strategies
• Token counting with tiktoken (local) or Anthropic API (async)
• Extensible pricing registry pattern

Usage

Calculate Costs

import { CostRegistry } from '@agent-qa/cost-registry';

const registry = CostRegistry.default();

const cost = registry.calculate({
  source: 'llm',
  provider: 'anthropic',
  model: 'claude-3-5-haiku-latest',
  usage: { inputTokens: 1000, outputTokens: 100 },
});

console.log(`Cost: $${cost.totalCost.toFixed(4)}`);

Count Tokens

import { createTokenCounter } from '@agent-qa/cost-registry';

// Count tokens locally with tiktoken
const counter = createTokenCounter();
const tokens = counter('Hello, world!');

// Count tokens via Anthropic API
const asyncCounter = createTokenCounter({ model: 'claude-sonnet-4-5' });
const count = await asyncCounter('Hello, world!');

Subpath Exports

| Path | Description | |------|-------------| | @agent-qa/cost-registry | Main exports | | @agent-qa/cost-registry/rates | Pricing strategies | | @agent-qa/cost-registry/counting | Token counters |

Supported Models

Anthropic:

• claude-opus-4-5, claude-sonnet-4-5
• claude-3-5-sonnet, claude-3-5-haiku
• claude-3-opus, claude-3-haiku

OpenAI:

• gpt-4o, gpt-4o-mini
• gpt-4-turbo, gpt-4
• gpt-3.5-turbo

API Reference

CostRegistry

class CostRegistry {
  static default(): CostRegistry;
  calculate(input: CostCalculationInput): CostResult;
}

Pricing Strategies

import {
  createPricingStrategy,
  anthropicPricing,
  openaiPricing,
  ANTHROPIC_PRICING,
  OPENAI_PRICING,
} from '@agent-qa/cost-registry/rates';

Token Counters

import {
  createTokenCounter,
  createAnthropicCounter,
} from '@agent-qa/cost-registry/counting';

License

MIT