What is PINTI?

PINTI (Payment Intent & Trust Interceptor) is an agentic spend policy layer. Standards decide how agents pay — PINTI decides whether they're allowed to.

The Problem

AI agents are starting to spend money — buying API credits, provisioning infrastructure, purchasing services. But there's no universal control layer that works across payment rails. PINTI fills that gap.

What PINTI Does

Deny / Approve / Allow — Every spend intent goes through your policy engine. Blocked merchants get denied, large amounts require human approval, everything else flows through.
Full Audit Trail — Every spend request is logged with the initial decision, final decision, agent ID, merchant, amount, and reason. Filter and search your logs anytime.
Centralized Policy — Define rules once, enforce everywhere. Per-transaction limits, daily/monthly caps, merchant allow/blocklists, approval thresholds.

How It Works

One API call. Your agent asks. PINTI decides.

1. Agent sends a spend intent

POST /api/v1/spend/evaluate
x-api-key: pinti_xxxxxxxx_...

{
  "agentId": "my-agent",
  "amountMinor": 3000,
  "unit": "USD",
  "merchant": "openai.com",
  "category": "api",
  "reason": "Monthly API credits"
}

2. PINTI evaluates and returns a decision

{
  "decision": "ALLOW",
  "decisionReason": "OK",
  "spendRequestId": "cm4abc123..."
}

Three possible decisions: ALLOW, DENY, or REQUIRE_APPROVAL.

Key Concepts

Concept	Description
Workspace	Your organization container. All policies, keys, handles, and logs belong to a workspace.
Policy	A set of rules: transaction limits, daily/monthly caps, merchant/category allow/blocklists, approval thresholds.
Payment Handle	The execution identity — SDK Guard (your agent pays), External Executor (your backend pays), or Managed (PINTI pays).
Spend Request	A record of an evaluation — the intent, the decision, and the reason. Your audit trail.
Approval	When a spend requires human sign-off, an approval is created. See Approvals & Callbacks.

Integration Options

PINTI works with any stack. Choose the integration that fits your setup:

SDK (@pinti/guard) — Type-safe Node.js/TypeScript client for server-side evaluation.
MCP Server — Model Context Protocol integration for Claude Desktop and AI assistants.
OpenClaw Skill — Plugin for the OpenClaw agent framework.
REST API — Direct HTTP calls from any language or platform.

Tip

New to PINTI? Start with the Quickstart guide to go from zero to your first spend evaluation in 5 minutes.