Go on.

On your terms.

AllowIt lets your AI agents spend and invest,
within a policy you set.

Give it
a policy.

What it may buy, how much it may spend, and until when.

TOOLS & DATA$10030 DAYS

Every request
is checked.

Inside the policy, the agent goes ahead.
Outside it, the request is refused.

You still
call it.

Unclear requests come back to you.
Revoke the policy, and it stops.

Give an agent a scope, a budget and an expiry. AllowIt checks every action against those terms. Outside the terms, nothing moves.

Follow the story ↓

A short history of handing over money

People have always handed money to others, with limits.

A cheque is an instruction to someone else: pay this person, this sum, from my account. The oldest surviving English one was written in 1659.

Messrs Morris & Clayton16 February 1659

Pay unto Mr Delboe

the sum of four hundred pounds

£400Nicholas Vanacker

Cards let others charge you, within limits a network enforces.

Diners Club let restaurants bill you without cash changing hands. Every card since works the same way: someone else takes the payment, and a network checks it first.

Charge cardMember since 1950Every charge checked by the network

The web was built without a way to pay.

When HTTP/1.1 was written, its authors reserved a status code for paying: 402 Payment Required. Its whole definition was one line: reserved for future use.

Request for Comments: 2068          January 1997

     Hypertext Transfer Protocol -- HTTP/1.1

10.4.3 402 Payment Required

   This code is reserved for future use.

Money became software.

Dollars became stablecoins that software can hold and send. Government debt followed, as tokenised treasury funds. Value could finally move at the speed of code.

USDCA digital dollar that software can hold and send
2018
BUIDLA tokenised fund of US Treasury bills
2024

Software learned to think, then to act and pay.

ChatGPT arrived in November 2022. Within three years agents were researching, writing code and running workflows on their own, and x402 and AP2 let them pay as they go.

2022 Here is a summary of the annual report.

2024 I compared five reports and drafted the analysis.

2025 I bought the missing dataset for 0.80 USDC and finished the analysis.

Agents are about to become first-class users of money.

The people who run the rails expect agents to make most online transactions. Forecasts for agent-led commerce run into trillions of dollars by the end of the decade.

$3–5trillion

Agent-led commerce worldwide by 2030, as forecast by McKinsey.

“In the not-too-distant future agents will account for most transactions online.”Patrick Collison, Stripe, April 2026

So why isn’t it happening yet?Nobody trusts an agent with money.

The rails exist. The money is on-chain. The agents are capable. But every era that handed money over had an instrument to bound it. Agents don’t have one yet.

1659Cheque

Pay this person, this sum.

1950Card

This card, checked on every charge.

Agents?

This agent, these actions, this much, until then. Ask me when unsure.

AllowIt gives agents that instrument: a policy.

Tell AllowIt what an agent may do with your money, in your own words. It writes the policy, hands it to your agent, and checks every action before money moves.

1659Cheque

Pay this person, this sum.

1950Card

This card, checked on every charge.

AgentsPolicy

This agent, these actions, this much, until then. Ask me when unsure.

See how it works

People will hand an agent the task.
Not yet the money.

Payments

23%trust AI to handle payments on their behalf

72%of US consumers have used an AI assistant

Visa Trust Index, Harris Poll, May 2026, 2,065 US consumers
Investing

5%give AI the authority to make the decision

68%of asset managers use AI in the investment process

Mercer, February 2026, 131 asset managers
Agent-led commerce grows only as fast as this gap closes.

Today there are two settings.
Neither one works.

Give it your keys.

It can buy, sell or send anything, to anyone, for any amount. You find out afterwards.

  1. Move 200 USDC into a treasury fundDone
  2. Buy a filings dataset · 0.80 USDCDone
  3. Rent two hours of compute · 3 USDCDone
  4. Rebalance, leaving 12% liquidYour floor is 20%Done. Nobody asked
  5. Shift 150 USDC into a peso fundCurrency risk unclearDone. Nobody asked
  6. Renew a data subscription · 9 USDCDone

6 done. 2 you would have refused.

Why wallets alone break when agents hire agents

One live budget, and a receipt for every step.

No two agents can spend the same dollar. Every receipt names the chain of authority behind it. Revoke a branch, and everything under it stops.

You Your wallet10,000 USDC Your wallet10,000 USDC Policy10,000 USDC
Lead agent Your keyeverything you own 0x4f…a12,000 USDC 2,000 USDCthis project · 30 days
Researcher 0x8c…02500 USDC 0x8c…02500 USDC 300 USDCdata only
Source finder 0x2a…11300 USDC 0x2a…11300 USDC 100 USDCfilings only
Summariser 0xb0…7d100 USDC 0xb0…7d100 USDC No spendingreads, never pays
Trading desk 0x1d…9e1,500 USDC 0x1d…9e1,500 USDC 1,500 USDCapproved assetsRevoked by you
Executor 0xe5…381,500 USDC 0xe5…381,500 USDC 1,500 USDCtrades up to 250
Data buyer 0x73…c4400 USDC 0x73…c4400 USDC 150 USDCdata APIs
Email checker 0x6f…d2200 USDC 0x6f…d2200 USDC 50 USDCHunter onlyPaid Hunter 0.03 USDC
Price watcher 0x9b…40400 USDC 0x9b…40400 USDC 20 USDCprice data onlyPrompt-injected · sent 400 USDCAsked for 400 · denied at 20

From a sentence
to a policy.

Run my research desk with up to 1,000 USDC. Keep at least 20% in USDC and no more than 25% in any one asset. You may buy and sell approved tokenised funds, and buy the data you need for the analysis. Ask me before anything with unusual currency risk. Stop at the end of the year.

Let it run the portfolio.
You set the policy.

Stablecoins, tokenised funds, on-chain positions. An agent can rebalance, buy and redeem around the clock, inside a policy you approve: which assets, how much per trade, what must stay liquid, and when it comes back to you.

Policy · Portfolio

1,000 USDC

Assets
Approved list only
Per trade
250 USDC
Cash floor
20% in USDC
Concentration
25% per asset
Judgement
No unusual currency risk

Policy created from this prompt: “Manage 1,000 USDC of approved assets. Keep 20% liquid, nothing above 25% in one asset, and ask me about unusual currency risk.”

Your agent tries to:

AllowIt checks it against the policy:

  1. Approved assetTreasury fundpasses
  2. Per trade200 of 250 USDCpasses
  3. Cash floor24% left · floor 20%passes
  4. Concentration18% · limit 25%passes
  5. Your judgementFit 0.91 · go at 0.85passes
Allow

Done.

Every rule passes. The agent goes ahead on its own.

You give it a job.
It should be able to finish.

Some of what an agent needs is not on the open web: proprietary data, specialised models, verified contacts. x402 lets it buy exactly that, one request at a time. Here is one job, with and without AllowIt.

You approve a $15 policy for this job.

It sets what the agent may buy, how much, and for how long. Your keys stay with you.

You

Find Berlin warehouses with the best rooftop solar potential and who decides on energy at each, with a verified work email. Make the top ten a one-page pitch showing panels on their own roof. Spend up to $15. Business data only.

Your agent

I can list the buildings, but roof data, contact databases, aerial images and image generation are all paid. I don’t have access to any of them. Can you set that up?

Solar prospectingApproved by you
$15one shared limit
  • Business data only
  • Images up to $0.25 each
  • Expires in 2 hours

AllowIt checks these terms before every payment.

The budget splits with the work.

Each agent gets a share of its parent’s allowance: the email checker $1 of the contact finder’s $7, the renderer $2.50 of the pitch designer’s $5. Every share comes out of your $15.

Your prospecting agent

Splits the work and hands each part a narrower allowance.

Site scout

Finds the best roofs

$3 ceiling · solar data

60 warehouses from open map data · free

Google Solar API
60 roofs analysed$1.20 paidSunshine and roof geometry come from Google’s model, not the open web

Contact finder

Who decides on energy

$7 ceiling · people data
FullEnrich
25 decision-maker searches$3.75 paidRoles and work contacts live in licensed databases
Delegates inside its $7

Email checker

$1 ceiling · Hunter only

Hunter
22 emails verified$0.66 paid · inside the $7Proving a mailbox exists needs a trusted mail server

Pitch designer

One page per prospect

$5 ceiling · images
Google Solar API
10 aerial roof images$0.80 paidHigh-resolution aerial imagery is licensed, not free
Delegates inside its $5

Renderer

$2.50 ceiling · image model only

Tried a premium image model at $0.50 an image. Refused: your limit is $0.25 an image.

Image model
10 panel renders$2.00 paid · inside the $5Drawing panels onto a real roof takes a paid model on a GPU

The results return to your prospecting agent.

$8.41 spent · $6.59 left · 1 request refused

Every payment is on record.

The log shows what was bought, by which agent and for how much, including the request that was refused.

Your agent

Your solar prospects are ready.

22 warehouses with strong rooftop solar potential, the person who decides at each and a verified work email, plus a one-page pitch for the top ten with panels drawn on their own roof. I spent $8.41.

Payments · solar prospects5 approved · 1 refused
  1. ApprovedGoogle Solar API60 rooftop solar analysesSite scout$1.20
  2. ApprovedFullEnrich25 decision-maker searchesContact finder$3.75
  3. ApprovedHunter22 email verificationsEmail checker$0.66
  4. ApprovedGoogle Solar API10 aerial roof imagesPitch designer$0.80
  5. RefusedPremium image model10 renders at $0.50 each · over your $0.25 limitRenderer$0.00
  6. ApprovedImage model10 panel renders at $0.20 eachRenderer$2.00
Paid$8.41 of $15

Unit prices from the x402 catalogs of StableEnrich (data) and StudioMCPHub (image generation), October 2026.

You set the bounds.
Your agent chooses the actions.

Try it in the demo: write a policy in plain words, hand it to an agent, and watch every action get checked.

  1. 01Open the workspace

    In the header, set the network to Local dev. No wallet, no funds.

  2. 02Write a policy

    Pick an example or describe your own, in plain words. Generate the policy.

  3. 03Hand it over

    Approve the budget to create the agent skill, then press Start agent.

  4. 04Watch it work

    Follow each request and decision live. Answer the ones that come back to you.

Open the demo workspace

About five minutes. Runs on mock execution: every decision is real, no money moves.

Common questions.

What is a policy?

The rules an agent must follow when it touches your money: what it may do, with which assets or services, how much, until when, when it must ask you, and what it may hand on to other agents. AllowIt writes it from your words. You approve it before it takes effect.

What is Ackrate?

Ackrate is our policy engine. It checks each proposed action against your policy before execution, outside the agent, and records the decision.

Does the agent get my money or my keys?

No. Your agent receives a skill and an access token scoped to one policy. Access expires, and cancelling the policy revokes it. AllowIt enforces the policy outside the agent; the agent’s tools cannot change or bypass it.

What happens when an action does not fit?

It is refused, with the reason. A hard limit, such as a per-trade cap, cannot be overridden. Actions that pass the limits but leave the judgement unclear wait for your answer.

Which agents can use it?

Any agent that can read a skill and use a tool: hand it the skill AllowIt creates for your policy. Or let the hosted agent in AllowIt run the job for you.

Does real money move in the demo?

No. The demo records each approved action against your budget and uses mock transfers on Solana and Stellar. No wallet connects and no funds move.