@letagentpay/ai

LetAgentPay tools for Vercel AI SDK — spending governance for AI agents.

Installation

npm install @letagentpay/ai ai zod

Quick Start

import { generateText } from "ai";
import { anthropic } from "@ai-sdk/anthropic";
import { createLetAgentPayTools } from "@letagentpay/ai";

const tools = createLetAgentPayTools({ token: "agt_..." });

const { text } = await generateText({
  model: anthropic("claude-sonnet-4-20250514"),
  prompt: "Buy the cheapest weather API plan",
  tools,
});

Tools

| Tool | Description | | --- | --- | | requestPurchase | Request approval to spend money (call BEFORE any purchase) | | checkBudget | Check current budget, spent, held, and remaining | | listCategories | List valid spending categories | | myRequests | List recent purchase requests (optionally filtered by status) | | confirmPurchase | Confirm or report failure after completing a purchase |

Configuration

// Token from parameter
const tools = createLetAgentPayTools({ token: "agt_..." });

// Token from LETAGENTPAY_TOKEN env var
const tools = createLetAgentPayTools();

// Self-hosted instance
const tools = createLetAgentPayTools({
  token: "agt_...",
  baseUrl: "https://your-instance.com/api/v1/agent-api",
});

Using Individual Tools

import { createRequestPurchaseTool, createCheckBudgetTool } from "@letagentpay/ai";
import { LetAgentPay } from "letagentpay";

const client = new LetAgentPay({ token: "agt_..." });

const tools = {
  purchase: createRequestPurchaseTool(client),
  budget: createCheckBudgetTool(client),
};

How It Works

1. Agent decides it needs to make a purchase
2. It calls requestPurchase with amount, category, merchant, and description
3. LetAgentPay checks the request against spending policies (budget limits, category restrictions, daily/weekly/monthly caps, schedule)
4. Tool returns the decision: auto_approved, pending (needs human review), or rejected
5. Agent proceeds only if approved, adapts if rejected

Streaming

Works with streamText the same way:

import { streamText } from "ai";

const result = streamText({
  model: anthropic("claude-sonnet-4-20250514"),
  prompt: "Find and buy a domain name under $15",
  tools,
  maxSteps: 5,
});

for await (const part of result.textStream) {
  process.stdout.write(part);
}

Resources

• LetAgentPay Documentation
• Agent API Reference
• Vercel AI SDK Docs