Paygente
Sandbox

Paygente

The authorization layer for AI agents.

Give agents permission to spend - without giving up control.

Payment rails move value. Paygente decides whether an agent may cause it to move.

Open Sandbox Connect an Agent

Public sandbox. Simulated payments. No real money moves.

Research Agent asks

Paygente answers

  • Model API, analysis run

    €4.62

    ALLOWED

    Approved provider, within the daily limit.

  • Dataset licence, 12 months

    €320.00

    NEEDS APPROVAL

    Above €250, so a person decides.

  • Unlisted data vendor

    €18.00

    REFUSED

    Not an approved recipient.

Illustrative. Try a live decision in the sandbox.

How it works

The agent asks. Your policy answers.

  1. 01

    Agent

    Asks before it acts: what it wants to buy, from whom, for how much, and why.

  2. 02

    Paygente

    Identifies the agent from its key and loads the policy that governs it.

  3. 03

    Policy

    Limits, approved recipients, approval thresholds and expiry, evaluated deterministically.

  4. 04

    Decision

    AllowApprovalDeny

    Allow, require a person’s approval, or deny - with every reason, in the same response.

  5. 05

    Payment rail

    Moves the value, once authorised. The rail is not Paygente; in the sandbox it is simulated.

Payment rails move value. Paygente decides whether an agent may cause it to move - and records why.

For AI agents

Ask before you spend. Get the answer in one call.

An agent requests an economic action. Paygente returns a policy decision - allow, require approval or deny - with its reasons, in the same response.

MCP
Connect through MCP with npx -y @paygente/mcp-server and an agent key. MCP server
REST
The same decision over HTTPS, described by an OpenAPI contract. openapi.json
For models
A plain-text summary of what Paygente is and what an agent may do. llms.txt
  • The agent never receives a payment credential. It receives a decision.
  • An agent cannot approve, execute or revoke anything - including its own requests.
  • Its identity comes from its key, resolved on the server. Nothing it sends can widen it.
Connect an Agent Read the docs
Spend request and decision
POST /api/v1/spend-intents
Authorization: Bearer <agent key>

{
  "agentId": "agt_01JQ8F7Z9K3M4N5P6Q7R8S9T0V",
  "recipientId": "rcp_01JQ8F7Z9K3M4N5P6Q7R8S9T0V",
  "asset": "EUR",
  "amount": "1180.00",
  "category": "software",
  "commitmentType": "PURCHASE_ORDER",
  "purpose": "Observability plan, 12 months.",
  "projectReference": "CC-OPS-114"
}

→ REQUIRE_APPROVAL
  AMOUNT_REQUIRES_APPROVAL
  "€1,180.00 is at or above the €250.00
   threshold, so a person has to approve it."

For humans

What people do

  • Set the rules.

    Define what agents can spend and where.

  • Approve exceptions.

    Step in when policy requires human authorization.

  • See what agents did.

    Review decisions and activity.

Controls

Enforced on the server, not in the prompt.

Limits that add up

Per action, per day and per month. A retry loop that passes every single check still meets the running total.

Approved recipients

An agent spends only with recipients a person approved.

Approval thresholds

Above an amount you choose, a person decides. The requester cannot be the approver.

Expiry

Authority lapses. An expired authorization cannot execute, even after approval.

Revocation and kill switch

Withdraw one authorization, or stop every execution in the organization at once.

Audit trail

Every request, decision and approval, with its reasons. Append-only.

Security and boundaries

The agent proposes. Software decides.

Paygente is

  • An authorization layer: it decides and records.
  • Deterministic: no language model takes part in a decision.
  • Isolated per organization, with an append-only audit trail.

Paygente is not

  • A payment rail, a bank, a wallet or a payment processor.
  • Holding funds. No funds are held, no card is issued, and no payment provider is contacted.
  • Licensed or regulated, and it claims to be neither.

Today your integration enforces the decision: it asks Paygente and respects the answer. A system that never asks is not stopped by Paygente.

Security and threat model

Give agents permission to spend - without giving up control.

Open Sandbox

No account. Simulated payments. No real money moves.