AI Spend Guard

Stop your AI calls from wasting money.

The Problem

Running AI calls in production gets expensive fast. A single classification task might cost $0.66 with GPT-4o when GPT-4o-mini would've cost $0.02.

Without guardrails: $0.66
With Spend Guard: $0.02
Saved: 97%

Install

npm install ai-spend-guard

Quick start (policy only)

import { createSpendGuard } from "ai-spend-guard";

const guard = createSpendGuard({
  cacheRepeatedPrompts: true,
  autoReroute: true,
  maxCostPerCall: 0.2
});

const result = guard.evaluate({
  task: "classification",
  model: "gpt-4o",
  prompt: "Classify this support ticket",
  estimatedInputTokens: 180,
  estimatedOutputTokens: 40
});

console.log(result.action, result.model, result.savingsUsd);

Wrap live calls

import { createAIGuard } from "ai-spend-guard";

const guard = createAIGuard({
  apiKey: process.env.HUGGINGFACE_API_KEY!,
  provider: "huggingface",
  huggingfaceModel: "deepseek-ai/DeepSeek-R1:fastest",
  policy: { cacheRepeatedPrompts: true, autoReroute: true }
});

const response = await guard.chat({
  model: "deepseek-ai/DeepSeek-R1:fastest",
  messages: [{ role: "user", content: "Classify this ticket" }]
}, "classification");

console.log(response.choices?.[0]?.message?.content);

Pricing

Model pricing lives in pricing.rates.json. You can override it with the AI_SPEND_GUARD_PRICING environment variable.

Development scripts

• npm run check - type-check the project
• npm run build - build to dist/
• npm run demo - run the local demo

Usage Guide

See the usage guide for integration examples, policy examples, and a middleware sample: USAGE.md