Financial Authorization API

Real-time authorization for AI agent financial actions

Before your agent issues a refund, approves a payment, or applies a discount — SpendGuard checks the policy and returns allow, block, or escalate in under 50ms.

Try Live Demo Read the Docs

POST /v1/checks

{
  "agent_id": "support-agent-v2",
  "policy_id": "refund_policy",
  "action_type": "refund",
  "amount": 750.00,
  "currency": "USD",
  "counterparty": "customer_7821"
}

Response — 34ms

{
  "decision": "block",
  "reason_code": "max_amount_exceeded",
  "message": "Amount $750.00 exceeds the limit of $500.00.",
  "violated_rule_id": "r1",
  "next_step": "Reduce the amount or escalate to a manager."
}

How it works

Three steps from policy to decision.

Define your rules

Set dollar limits, vendor allowlists, time restrictions, geography blocks — 10 rule types to build any policy.

policy rule

{
  "rule_type": "max_amount",
  "parameters": { "limit": 500 }
}

Agent calls one endpoint

One API call before every financial action. Under 50ms. Your agent doesn't even notice.

API call

POST /v1/checks
{ "action_type": "refund",
  "amount": 600 }

Get a decision

Allow, block, or escalate. Every decision includes the rule that fired and a recommended next step.

response

{
  "decision": "block",
  "message": "Amount exceeds $500 limit"
}

Built for production

Everything you need to control AI agent financial actions.

10 Rule Types

Amount limits, vendor allowlists, time restrictions, geography blocks, discount caps, duplicate guards, and more.

Immutable Audit Log

Every decision logged with 14 fields. Append-only. Query by agent, action type, decision, or date range.

Policy Versioning

Every update creates a new version. Previous versions preserved. Checks log which version was used.

Duplicate Guard

SHA-256 fingerprint blocks repeated actions within configurable time windows. Stops agent retry loops.

Semantic Classification

Handles ambiguous inputs. "Make the customer whole" resolves to refund. The classifier resolves. Rules still decide.

MCP Ready

AI agents discover and use SpendGuard through Model Context Protocol. Works with Claude and any MCP client.

Try it live

This calls the real API. No signup required. No data stored.

AI Support Refund Policy Demo Mode

Demo temporarily unavailable. Try again in a moment.

$50 refund

Order placed 10 days ago

$750 refund

Over policy limit

$200 refund

Order placed 45 days ago

$300 refund

Above escalation threshold

$100 discount

15% off — within cap

This is a live API call to POST /v1/simulate. No data is stored. Try it with your own policies.

Simple, transparent pricing

Start free. Scale as you grow.

Free

For exploring and testing

$0 /month

1,000 checks/month
1 policy
Community support
Audit log access
Simulation endpoint

Start Free