The payment firewall for AI agents

The security layer between an agent deciding to spend and the money moving.

Try the live API

Try the live API

Get the SDK

Get the SDK

Get the SDK

Get the SDK

How it works

How it works

How it works

How it works

Task ↔ purchase

Does the purpose match the submitted task context?

Amount + budget

Is it within transaction, daily, and velocity limits?

Counterparty

Is the vendor or address blocked or otherwise risky?

PAYMENT SECURITY FOR AI AGENTS

Safe4 is the security layer between an agent deciding to spend and the money moving.

PAYMENT SECURITY FOR AI AGENTS

Safe4 is the security layer between an agent deciding to spend and the money moving.

PAYMENT SECURITY FOR AI AGENTS

Safe4 is the security layer between an agent deciding to spend and the money moving.

THE CHECKPOINT

Safe4 sits in front of execution. It checks the payment against the task that triggered it, the budget it draws from, and the counterparty receiving it.

THE CHECKPOINT

Safe4 sits in front of execution. It checks the payment against the task that triggered it, the budget it draws from, and the counterparty receiving it.

THE CHECKPOINT

Safe4 sits in front of execution. It checks the payment against the task that triggered it, the budget it draws from, and the counterparty receiving it.

THE DECISION

One authorization call evaluates policy, purpose, budget, velocity, counterparty, and scope — then allows or denies before a single cent settles.

THE DECISION

One authorization call evaluates policy, purpose, budget, velocity, counterparty, and scope — then allows or denies before a single cent settles.

THE DECISION

One authorization call evaluates policy, purpose, budget, velocity, counterparty, and scope — then allows or denies before a single cent settles.

Wallet controls check limits and addresses. They do not check whether the purchase matches the job.

ARCHITECTURE

One authorization call, one durable reason trail

Get the SDK

Get the SDK

Read the docs

Read the docs

ARCHITECTURE

One authorization call, one durable reason trail

Get the SDK

Get the SDK

Read the docs

Read the docs

ARCHITECTURE

One authorization call, one durable reason trail

Get the SDK

Get the SDK

Read the docs

Read the docs

Agent submits

The agent sends task context and budget allocation. Safe4 returns a 402 challenge stating price and required proof.

Task context

Budget allocation

402 challenge

Payment proof

Agent submits

The agent sends task context and budget allocation. Safe4 returns a 402 challenge stating price and required proof.

Task context

Budget allocation

402 challenge

Payment proof

Agent submits

The agent sends task context and budget allocation. Safe4 returns a 402 challenge stating price and required proof.

Task context

Budget allocation

402 challenge

Payment proof

Safe4 evaluates

Policy and purpose checks run in parallel: task ↔ purchase, budget, velocity, counterparty, autonomy scope, and payment proof.

Policy check

Purpose match

Scope validation

Audit trail

Safe4 evaluates

Policy and purpose checks run in parallel: task ↔ purchase, budget, velocity, counterparty, autonomy scope, and payment proof.

Policy check

Purpose match

Scope validation

Audit trail

Safe4 evaluates

Policy and purpose checks run in parallel: task ↔ purchase, budget, velocity, counterparty, autonomy scope, and payment proof.

Policy check

Purpose match

Scope validation

Audit trail

Wallet executes

On ALLOW, Circle Agent Wallet executes. On DENY, the executor is never invoked and no funds move.

ALLOW → execute

DENY → block

Circle integration

Zero false spend

Wallet executes

On ALLOW, Circle Agent Wallet executes. On DENY, the executor is never invoked and no funds move.

ALLOW → execute

DENY → block

Circle integration

Zero false spend

Wallet executes

On ALLOW, Circle Agent Wallet executes. On DENY, the executor is never invoked and no funds move.

ALLOW → execute

DENY → block

Circle integration

Zero false spend

Settlement verified

USDC settles on Arc with receipt verification. The decision and its evidence can be reconstructed later.

Arc settlement

USDC payment

Receipt verified

Full audit log

Settlement verified

USDC settles on Arc with receipt verification. The decision and its evidence can be reconstructed later.

Arc settlement

USDC payment

Receipt verified

Full audit log

Settlement verified

USDC settles on Arc with receipt verification. The decision and its evidence can be reconstructed later.

Arc settlement

USDC payment

Receipt verified

Full audit log

THE SECURITY GAP

The missing control point

Get the SDK

Get the SDK

Read the docs

Read the docs

THE SECURITY GAP

The missing control point

Get the SDK

Get the SDK

Read the docs

Read the docs

THE SECURITY GAP

The missing control point

Get the SDK

Get the SDK

Read the docs

Read the docs

A valid wallet action can still be the wrong action for the job. Same amount, same category, same counterparty — and still the wrong purchase. Safe4 inserts an enforceable checkpoint.

x402

Start with the SDK, or read how the policy engine reaches a decision.

Put a checkpoint in front of your agent's wallet

Get the SDK

Get the SDK

Read the docs

Read the docs

Put a checkpoint in front of your agent's wallet

Get the SDK

Get the SDK

Read the docs

Read the docs

/ Integrations

The rails and protocols Safe4 already speaks.

INTEGRATIONS

The rails and protocols Safe4 already speaks.

INTEGRATIONS

The rails and protocols Safe4 already speaks.

Safe4's own decision is sold over x402. An agent calls, receives a price list, pays in USDC, and gets an authorization back.

Arrows clustered in a target

01 x402

Payment protocol

Safe4 is callable as an MCP tool. Connecting and reading the price are free; the authorization call settles per use.

Model Context Protocol logo

02 Model Context Protocol

Agent tool

On ALLOW, Circle Agent Wallet executes. On DENY the executor is never invoked and no funds move.

Circle logo

03 Circle Agent Wallet

Execution

USDC settles on Base through open x402 facilitators, then is verified against the chain rather than trusted.

Base logo

04 Base

Settlement network

USDC settles on Arc with receipt verification, so the decision and the movement of money stay tied together.

Arc logo

05 Arc

Settlement network

Payments settle from a Gateway balance, so a buyer funds once and spends across many calls.

Circle logo

06 Circle Gateway

Settlement rail

Cataloged in the CDP facilitator's Bazaar, so an agent can find the endpoint and read its schema before paying.

Coinbase logo

07 x402 Bazaar

Discovery

Listed on the public x402 explorer, with the price and input schema readable by any agent that looks.

x402scan logo

08 x402scan

Discovery

Agent Payments Protocol mandates are verified before a payment is allowed to proceed.

Agent Payments Protocol logo

09 AP2

Mandate verification

COMMON QUESTIONS

Your most common questions answered

COMMON QUESTIONS

Your most common questions answered

COMMON QUESTIONS

Your most common questions answered

What is Safe4?

How does Safe4 work?

What does Safe4 evaluate?

How is Safe4 different from wallet controls?

What technologies does Safe4 use?

Is Safe4 a replacement for Circle Agent Wallet?

TALK TO US

Design partners and integration questions

TALK TO US

Design partners and integration questions

TALK TO US

Design partners and integration questions

Get the SDK

Start integrating Safe4 into your agent. Check the quickstart guide for setup instructions.

View Quickstart

View Quickstart

Read the Docs

Full SDK reference, policy engine documentation, and integration patterns.

View Docs

View Docs

View on GitHub

Source code, examples, and contribution guidelines. Open issues and PRs welcome.

Go to GitHub

Go to GitHub

Talk to us about your integration

Talk to us about your integration

The rails for autonomous commerce are landing now. The control layer above them is still open.