What is pocket agent

pocket agent gives you your own AI agent with a real card, wallet, and inbox — it shops, pays, and gets things done, while you stay in control.

Most AI assistants can talk about the world. A pocket agent can act in it. It has a card it can pay with, a multi‑currency wallet it can draw from, an email address it can use to sign up and read confirmations, and real hands — a live browser and an agentic checkout — to actually finish the job. You give it a goal in plain language; it does the research, the clicks, and the typing; and anything that costs money waits for your tap.

Be everywhere, live here. Your agent works across the web on your behalf — but everything it does flows back to one place you control.

The thesis

The web is still built for humans doing one tedious step at a time: open ten tabs, compare prices, create an account, copy a one‑time code out of your email, fill the same shipping address again, reach the checkout, pay. pocket agent collapses that into a single sentence and a single approval.

The product is organized around two ideas:

• Be everywhere. The agent can go wherever the task lives — a store, a flight search, a sign‑up form, your own inbox — using a real stealth browser and an agentic commerce rail. It isn't a chatbot pointing you at links; it completes the whole flow.
• Live here. Money, identity, memory, and approvals stay inside pocket agent. One card you can freeze, one wallet you can see, one inbox per agent, one approval sheet for anything irreversible.

The result is an assistant that is genuinely useful and safe to let near real money, because the money path is narrow, guard‑railed, and always ends with you.

What an agent is

An agent is a specialist worker that belongs to you. Unlike a single all‑purpose chatbot, each agent is a self‑contained unit with its own resources, so you can keep concerns cleanly separated — a shopping agent, a travel agent, a bills agent — each with its own budget and history.

Every agent carries:

It has…	So it can…
Its own card	Pay for things, within limits you set
Its own wallet	Hold dollars, euros, and crypto across chains
Its own inbox	Sign up for sites, read OTPs, send and reply to email
Its own memory	Remember your sizes, address, brands, and budgets
Its own tasks	Run standing jobs on a schedule and ping you
Real hands	Browse the live web and complete real checkouts

You create an agent by giving it a name and picking a role. The role is a server‑owned persona — it shapes how the agent works and sets sensible default spending guardrails, but it can never override the safety rules.

Role	Focuses on	Default per‑tx / daily / approve‑over
Personal	Everyday research, planning, writing, reminders, the occasional buy	$200 / $400 / $100
Shopping	Finding products, comparing prices and reviews, tracking deals, buying	$300 / $600 / $100
Travel	Flights, hotels, itineraries — books when it's genuinely cheap	$1500 / $3000 / $500
Finance	Balances, budgeting, market research — conservative, allowlist merchants, asks before spending	$200 / $500 / $50

Under the hood the agent runs a tool‑using loop (its "brain") that can search and open the web, pay with its card, shop through an agentic‑commerce catalog, read and send email, manage its wallet, and remember durable facts. It replies in your language (EN/RU today) and formats its answers clearly.

What it can do

A pocket agent is a general‑purpose assistant first, with real‑world abilities bolted on. Concretely, it can:

• Research and compare. Search the web, open pages, and read a real JS‑rendered browser — not just fetch text. It handles bot walls and CAPTCHAs by retrying and re‑opening blocked pages in a stealth browser.
• Shop and buy. Find a product, pick the size and color, get a quote, and complete the whole order through an agentic‑commerce rail. You tap Approve & pay once (passkey / Face ID) — the agent never hands the checkout back to you half‑finished.
• Pay with a card. Charge its virtual Visa within your limits. Every charge declares a policy hold first, so an unexpected charge is declined, not paid.
• Handle its own email. Create an alias, sign up for sites, pull the one‑time code out of its inbox, and send replies or confirmations.
• Move money. See balances across Ethereum, Base, BNB, and Solana; show a deposit address; propose a withdrawal, swap, or transfer for you to approve.
• Run on a schedule. Watch a price and buy under a threshold, draft a morning brief, guard your subscriptions, watch a flight route — standing tasks that run on their own and notify you.
• Remember and remind. Keep durable facts (address, sizes, budget) and ping you at a set time.

You always have the final say

Every action that spends money or is otherwise irreversible — a card payment, a withdrawal, a swap — is propose‑only: the agent prepares it and you approve it. Autonomous background runs are even stricter: irreversible tools are hidden and hard‑blocked, so a scheduled task can research and alert, but it can never move money without you. See Security & control.

Some capabilities are plan‑ or config‑gated, and the app says so plainly when they are:

• Stealth browser and voice input are available on paid plans.
• Agentic DeFi (swap, bridge, stake, lend) appears only when it's configured, and every DeFi action is still approval‑gated.
• Add funds from a bank (card / Apple Pay on‑ramp) and cash out to a bank are shown as coming soon until enabled; funding and withdrawing in stablecoins work today.
• A real phone number for your agent (calls and SMS/OTP) is coming soon.

The three pillars

Three things turn a clever chatbot into an agent that can actually finish errands. Each agent gets all three.

💳 Card — so it can pay

Each agent can carry a virtual Visa (USD). You fund it from your wallet balance (crypto, or an aggregate of your USD/EUR stablecoins across chains, cheapest chains first), set per‑transaction and daily limits, choose ask‑to‑approve over $X, and restrict merchants to any or an allowlist. You can freeze, reveal the full number, redeem the balance back to USD, or hand the agent your own card through a secure hosted page where the card details never touch the app. See Cards.

🪙 Wallet — so it has money to work with

A multi‑currency, multi‑chain wallet. The Currencies view shows USD and EUR as friendly fiat balances over stablecoins; Digital assets shows the raw on‑chain tokens across Ethereum, Base, BNB, and Solana. Sends are gasless on Solana and, where a sponsorship policy exists, on EVM too. You can deposit stablecoins, swap across chains, transfer to another user or agent, and read a clean on‑chain receipt for every move. See Wallet.

📥 Inbox — so it can act like a person online

Every agent can claim its own real email address. That's what lets it sign up for a store, receive the confirmation, and pull the OTP it needs to finish a purchase — then send you (or a merchant) a reply. You choose the alias; email isn't minted automatically. See Inbox & email.

The positioning in one line: your agent has a card, an inbox, crypto, and real hands to get things done.

Who it's for

pocket agent is for people who would happily delegate the boring, multi‑step parts of life on the internet:

• Busy delegators who'd rather say "just handle it" than open ten tabs.
• Deal hunters who want something watched and bought the moment it drops under a price.
• Travelers who want fares tracked and booked when they're genuinely cheap.
• Crypto‑native users who want a wallet and a spendable card in the same place, across chains.
• Inbox‑overwhelmed people who want renewals caught and a morning brief written for them.

It's not a licensed financial advisor and won't give personalized investment advice, and it deliberately keeps the money path narrow and approval‑gated — which is exactly what makes handing it a card reasonable.

A day in the life

Here's what a single day with a couple of agents looks like. Each example is one sentence from you and one approval — the agent does the middle.

Morning — the inbox gets triaged

Your bills agent has been watching overnight. It flags that two subscriptions renew this week, drafts a three‑line brief of what matters, and asks whether to cancel one. You tap once.

Midday — lunch, handled

"Order my usual lunch for 1pm." The agent remembers the place and your order, builds the whole cart, and shows you Approve & pay. One tap and it's on the way.

Afternoon — the sneakers you've been watching

You asked a shopping agent last week to watch a pair of sneakers and buy them under $120. They dropped. The agent completed the order and sends you the receipt — the approval you gave when you set the limit was enough.

Evening — groceries restocked

"We're out of coffee and oat milk — reorder." The agent rebuilds the basket from what you bought last time, stays under the daily limit, and files the receipt in its inbox.

How that one errand actually flows

Behind each of those taps, the loop is the same: you ask, the agent works, and the single moment that touches money comes back to you.

The core loop

Every task — one‑off or scheduled — follows the same shape. You stay at both ends of it.

The pattern never changes: you ask → your agent works → errands, payments, and tasks get done → you approve anything that matters → it reports back. Everything reversible, the agent just does. Everything that spends or can't be undone waits for your tap. Autonomous runs that go off on their own can research and notify, but are hard‑blocked from touching money.

Where to go next

• Quickstart — sign in and finish your first task in a few minutes.
• Agents — roles, memory, and how to talk to an agent.
• Autonomous tasks — standing jobs that run on a schedule.
• Cards · Wallet · Inbox & email — the three pillars in detail.
• Security & control — spend limits, approvals, and why the money path is safe.
• Platforms — web, PWA, Telegram Mini App, and native iOS/Android.