Developer Documentation

Quick Start

1. Sign up with your email (magic link, no password)
2. Create an agent in the dashboard and get your Bearer token (agt_...)
3. Choose your integration method below and start sending requests

Authentication

All Agent API endpoints use Bearer token authentication:

Authorization: Bearer agt_your_token_here

Integration

pip install letagentpay

from letagentpay import LetAgentPay

client = LetAgentPay(token="agt_your_token")

# Submit a purchase request
result = client.purchase(
    amount=49.99,
    category="groceries",
    merchant_name="Whole Foods",
    description="Weekly groceries",
)

if result.status == "auto_approved":
    client.confirm(result.request_id, success=True, actual_amount=47.50)
elif result.status == "pending":
    final = client.wait_for_decision(result.request_id, timeout=300)
    if final.status == "approved":
        client.confirm(result.request_id, success=True)
elif result.status == "rejected":
    print(f"Rejected: {result.policy_check}")

from letagentpay import LetAgentPay, guard

client = LetAgentPay(token="agt_your_token")

@guard(client, category="groceries")
def buy_groceries(items: list[str]) -> dict:
    total = calculate_total(items)
    return {"items": items, "total": total}

Framework Integrations

LetAgentPay works with popular AI agent frameworks. Each integration wraps the Python SDK as a native tool for the framework.

from letagentpay_langchain import LetAgentPayTool

tool = LetAgentPayTool()
agent = create_tool_calling_agent(llm, [tool], prompt)

@function_tool
def request_purchase(amount, category, ...):
    return client.request_purchase(...)

@tool("Request Purchase")
def request_purchase(amount, category, ...):
    return client.request_purchase(...)

"letagentpay": {
  "command": "npx",
  "args": ["letagentpay-mcp"],
  "env": {"LETAGENTPAY_TOKEN": "agt_..."}
}

Base URL: https://letagentpay.com/api/v1/agent-api

Request Lifecycle

1. Agent sends POST /requests with amount, category, description
2. Policy engine runs checks: status, category, per-request limit, schedule, daily/weekly/monthly limits, budget, account budget rules
3. All checks pass + auto-approve criteria met → auto_approved
4. All checks pass, no auto-approve → pending (human reviews)
5. Any check fails → rejected (with detailed explanation)
6. Pending requests expire after 30 minutes if not reviewed
7. After approval, agent confirms with POST /requests/{id}/confirm