About LetAgentPay

What is LetAgentPay?

LetAgentPay is a policy middleware for AI agents. It sits between your AI agent and any financial transaction, acting as a checkpoint that enforces the spending rules you define. Every time an agent wants to spend money — whether calling a paid API, provisioning cloud resources, or purchasing a product — it must first request approval from LetAgentPay. Only if the request passes your policy checks does the agent proceed to pay.

We are not a payment processor. We don't store credit card numbers, handle transactions, or move money. We are the control layer that decides whether a payment should happen — based on rules you set.

The problem

AI agents are becoming autonomous. They browse the web, call APIs, manage infrastructure, and make purchasing decisions. But the tools that give agents these capabilities don't come with spending controls built in.

Today, if you want to limit what your agent spends, you have two options: write it into the prompt ("don't spend more than $50") or hardcode limits in your application. Prompt instructions are unreliable — agents can misinterpret or ignore them. Custom code works but is fragile, hard to maintain, and doesn't generalize across agents or teams.

The result: unexpected charges, budget overruns, and no clear audit trail of what was spent, when, and why.

How it works

You create an agent in LetAgentPay and define its spending policy. Policies can be written in plain English — for example, "maximum $200/month on cloud services, no spending on weekends, individual purchases under $50 are auto-approved" — and our AI converts them into structured, enforceable rules. Or you can write them directly as JSON.

Each agent gets a unique Bearer token. When your AI agent wants to make a purchase, it calls our API (or uses the SDK or MCP server) with the amount, category, merchant, and a short description. The request goes through 8 policy checks:

1. Agent status — is the agent active?
2. Category — is this spending category allowed?
3. Per-request limit — does this single purchase exceed the cap?
4. Schedule — is spending allowed right now (day, time)?
5. Daily limit — would this push today's spending over the daily cap?
6. Weekly limit — same check for the current week
7. Monthly limit — same check for the current month
8. Total budget — would this exceed the agent's overall budget?

If all checks pass and the request meets your auto-approve criteria, it's approved instantly — the agent can proceed with the purchase. If it passes but doesn't qualify for auto-approval, it enters a "pending" state and you receive a notification to review it manually. If any check fails, the request is rejected immediately with a detailed explanation of which rule was violated.

After a purchase is completed, the agent sends a confirmation back with the actual amount spent. Spending counters update in real time, so subsequent requests are checked against accurate, up-to-date numbers.

Where it fits in the payment lifecycle

A typical AI agent payment flow looks like this:

LetAgentPay operates at steps 2 and 4 — the pre-payment authorization and post-payment confirmation. We don't touch the actual payment infrastructure. Your agent uses whatever payment method it already has (API keys, corporate cards, cloud billing). We add the policy layer on top.

This design means there's no vendor lock-in and no need to change how your agent pays for things. You just add one checkpoint before and one confirmation after.

What you can control

• Budgets — total budget per agent, plus daily, weekly, and monthly spending limits. When a limit is reached, new requests are automatically rejected.
• Categories — define which spending categories an agent can use (e.g., "cloud", "groceries", "saas"). Requests outside allowed categories are rejected.
• Per-request caps — maximum amount for a single purchase. Prevents large unexpected charges.
• Time schedules — restrict spending to specific days and hours. Block weekends, nights, or define custom windows.
• Auto-approval rules — define conditions under which purchases are approved without your manual review (e.g., "auto-approve if under $20 in allowed categories").
• Natural language policies — write rules in plain English. Our AI converts them to structured JSON policies that the engine can enforce deterministically.

Who is it for?

How to integrate

LetAgentPay offers three integration methods:

• Python SDK — install letagentpay and add a few lines of code. Supports both direct calls and a decorator pattern.
• MCP Server — works with Claude Desktop, Cursor, and any MCP-compatible tool. One configuration block and your agent gets spending tools automatically. Best for non-technical users.
• REST API — simple HTTP endpoints with Bearer token authentication. Works with any language or framework.

See the developer documentation for code examples, API reference, and the full request lifecycle.

Design principles

• Pre-payment, not post-payment. Approval happens before the money moves. You don't find out after the fact.
• Deterministic enforcement. Policies are converted to structured rules and checked programmatically. No prompt-based "hoping the agent listens."
• Human in the loop. Routine purchases flow through automatically. Unusual ones surface to you for review. You choose the threshold.
• Full transparency. Every request, every check result, every decision is logged. Complete audit trail from day one.
• No lock-in. We don't touch your payment methods. Add one API call before payment and one after — that's the entire integration surface.

Contact us

Found a bug? Have a feature request? Something doesn't work as expected? We want to hear from you — your feedback helps us build a better product.

Drop us a line at support@letagentpay.com. We read every message and typically respond within a day.