mandatumio 1.0.0

Zero-trust authorization and spending limits for AI agents. Ed25519-signed JWTs, policy enforcement, and payment gating.

mandatum

Spending limits and authorization for AI agents. One line of code to prevent your AI from overspending.

Install

pip install mandatum

Quick Start

from mandatum import Mandatum

m = Mandatum(
    api_key="mdt_live_...",
    base_url="https://mandatum-fbg9.onrender.com"
)

# Register an agent
agent = m.agents.create("ShoppingBot", type="payment", delegator="[e-mail]")

# Check spending limits before payment
result = m.agents.spend(agent["id"], {
    "amount": 350,
    "currency": "INR",
    "vendor": "DigitalOcean",
})

if result["allowed"]:
    razorpay.order.create({"amount": 35000})
else:
    print("Blocked:", result["reason"])

Payment Wrappers

from mandatum.payments import MandatumRazorpay

pay = MandatumRazorpay(razorpay_client, m, agent["id"])
result = pay.create_order(amount=50000, currency="INR")

if result["allowed"]:
    print("✅ Order created:", result["order"])

Decorator Guard

from mandatum.integrations import mandatum_guard

@mandatum_guard(m, agent_id="agt_abc123", action="payments:execute")
def buy_groceries(items, amount=0):
    return razorpay.order.create({"amount": amount * 100})

# Automatically checks limits — raises MandatumBlocked if over
buy_groceries(["milk", "bread"], amount=500)

Context Manager

from mandatum.integrations import MandatumGuard

with MandatumGuard(m, "agt_abc123") as guard:
    guard.check_spend(amount=350, vendor="AWS")
    # Only runs if spend was allowed
    aws.create_instance(...)

API Reference

Method	Description
m.check(token, action, context)	1-line permission gate
m.agents.spend(id, params)	Record spending + check limits
m.agents.spending(id)	Get spending summary
m.agents.create/list/get/suspend/revoke	Manage agents
m.tokens.issue/verify/revoke	Manage tokens
m.policies.create/list/assign	Manage policies
m.audit.list/verify_chain	Audit trail

License

MIT