What it does
What is account abstraction, and do I need to understand it?
No, you do not need to understand it. It is the reason any of this works. An ordinary wallet is an address controlled by one private key: every transaction is signed by that key, and anything holding the key can do anything. Account abstraction makes the account a contract instead, so the rules about who may do what are code. That is what lets your account give permission to an agent contract instead of handing out a key, and refuse a plain key outright. It is what lets a permission be scoped to named operations with a daily limit per token, instead of all-or-nothing. And it is why somebody else can pay your network fees, so you never have to hold the native coin of a network you have not used. The standards are ERC-4337 for the account and ERC-7579 for the permission modules. You can use the product without ever meeting those numbers. They are named here because they are where the claim “the agent cannot hold your money” is enforced in code.What is an agentic wallet, and is this one?
An agentic wallet is an account an AI agent can spend from without asking you for each action, inside limits set in advance. Yes, this is one. The main design question is what enforces the limits. Three answers are in circulation: the agent holds a private key and is asked in software to behave; the agent holds part of a key, with a policy engine deciding what gets co-signed; or the agent holds no key over your account at all. This is the third. The agent is a contract your account gives permission to. Its signing key lives inside a hardware enclave and can only drive that contract. It can only pick an operation from a fixed set of actions and supply its details, such as the amount, asset and network. It cannot set a destination address, so an instruction that sends your money elsewhere cannot be written down, and nothing has to catch it. A daily limit per token is set on your account, off until you set it, and nothing on our side can raise it. “Non-custodial” is something you can test here: you can move everything out with your own wallet, with no cooperation from us and with our servers switched off.What is this?
An agent that turns a sentence into transactions, and then keeps running it. You write what you want in ordinary words. It works out which network, which market, which pool, what to approve, and what order it all has to happen in. Then it executes. Some of that is work you could do yourself, more slowly, across half a dozen browser tabs. Some of it you could not do at all, which the next answer covers. It runs on an account that belongs to you. Nothing in the system can hold your money on the way through. It does not need you present: a strategy card waits for a condition, a price or a clock, and a workflow can run now, on a schedule, or when a price arrives. → How a run worksWhy can I not just do this myself?
Because a lot of it cannot be done by hand at all. Take moving a loan to a cheaper protocol. To repay the old loan you need your collateral back. To get your collateral back you have to repay the loan. By hand, the collateral stays locked until the loan is repaid, so your only way through is to find the entire loan amount somewhere else first. Here it is one instruction. The money to repay the old loan is borrowed and repaid inside the same transaction. Either the whole thing happens or none of it does, and your account never holds the borrowed amount. Three more things follow from the same idea:- Steps that can share a transaction are merged into one. “Give permission, then supply” as a single transaction means the permission cannot outlive a supply that failed and sit there afterwards. It is also cheaper. On a real plan, three transactions became two, and the fee fell with them.
- A step can spend exactly what the step before it produced. Nobody knows how much a swap returns until it runs. By hand you wait for it to land, read the number, and type it into the next transaction. Here the next step reads it directly. This is off by default and turned on per workflow, because it changes how money moves.
- The order is worked out rather than taken from your list. Steps that do not depend on each other run at the same time instead of queueing.
I have never used DeFi. What do I actually need to know?
Four words, and the rest of this page uses them.
You do not need to understand blockchains to read the rest. Where something
below is technical, it is there because the answer would be a half-truth without
it.
What can it actually do today?
It can run 80 operations across 11 protocols and venues on 4 networks. The catalogue lists them all. It can lend and borrow, provide liquidity, and borrow against a liquidity position to make it larger. It can swap one asset for another, move a loan from one market to a better one, and move USDC and WETH between networks. It can buy and sell property tokens and collect rent, and buy and sell Coinbase’s tokenized shares on Base. There is also a set of one-press exits. It also trades perpetuals at venues you connect. On Hyperliquid and Lighter, the agent opens and closes positions itself, through a card, inside the account locked to that venue. You connect, deposit and approve the card. Binance connects in demo mode only. → Trading venues Buying an option is still yours to do. The system does not reach an options venue, so the step sits in the plan to keep the hedge in the right order beside the position it protects. Nothing waits for it. It also answers questions without doing anything, which is the half people miss. See the next question. Almost every operation runs on Ethereum, Base and Arbitrum. HyperEVM is for moving USDC and for trading on Hyperliquid. Buying and selling property, and the four rent operations, run on Base only. → NetworksA crossing lands in your account, not in your wallet. USDC moves between
Ethereum, Base, Arbitrum and HyperEVM, and WETH between Ethereum, Base and
Arbitrum. A crossing usually takes a couple of minutes, but that is an
expectation and not a promise. Up to 1% comes off the amount. Plans are built around the wait, so a
step on the far side cannot start early. → Moving between networks
Can it tell me things without spending anything?
Yes, and most of what it can do is this. It has 52 tools. Five of them spend money. The rest only read. You can ask what a price range would have collected day by day on every network and fee tier, how seventeen different ways of managing that position would have done over real history, what nine option structures cost today and cost at every roll since 2021, what perpetuals pay on four venues, who is being liquidated and where, what prediction markets and the rate curve expect, and whether a two-year signal says be in the market or out of it. None of that signs anything or moves anything. It reads market history that is recorded on a schedule, so asking the same question twice gives the same answer. Ask before you act, because the numbers that are easiest to find are usually the wrong ones. A pool’s advertised rate is the last 24 hours across the whole pool, while your money sits in a price range that earns nothing on any day the price leaves it. One real position collected 57.9% in fees over a year and still finished down 6.4%, because the two assets it held moved apart underneath it. Start at surveying ranges or pricing a hedge. The market history behind all of it is on what we hold.What does it cost?
The platform fee is a markup on the network fee and is charged in credits. It is
not a cut of the money a step moves. The other two are.
Network fees themselves are paid by a key of ours and reimbursed, so you never
have to hold the native coin of a network you have not used.
Does it keep running, or do I press go every time?
A strategy card keeps running. You approve it once and start it, and from then on it acts by itself inside the limits you signed. It waits for a condition, a price or a clock. For example, it can buy every Monday, buy if the price comes back to a level, or sell when an indicator turns. A workflow can also be started on a schedule, at a price, or when another workflow finishes. Nothing is agreed in advance. Each time one fires, the plan is rebuilt and re-priced against the market at that moment. A schedule set up in March does not carry March’s prices into June. Starting a card, and turning a schedule on, are each a deliberate press, for the same reason pressing Run is. From then on it spends without asking you again. → Strategy cards and Running it without youWhat is the actual sequence: what do I do?
Say what you want. A plan is written from it: the steps, in order, with the amounts spelled out. You read it and approve it. Then it runs, and it keeps running until it finishes or stops to ask you something. The plan is the record that counts, more than the conversation that produced it. It is what you approved, what ran, and what gets checked if anything is ever disputed. → QuickstartMy run stopped. Did it fail?
Probably not. Stopping to ask is a normal outcome. It retries very little on purpose: a fee bump for a transaction stuck behind a busy network, and a couple of attempts where the cause can clear on its own. Past that it stops and asks, because most real problems need a decision from you, and another identical attempt would not help. To carry on, tell it to continue. One instruction covers this, and it works out which of three things you meant by looking at the run:
You are not asked to choose, because with three near-identical “spend money”
buttons, people press the wrong one.
Can it wait for a price?
Yes. You point it at a workflow you have already saved and give it a price to wait for. It does nothing at all until the price arrives. It agrees to nothing while it waits. So it can rest for days without going stale, it does not block your other runs, and an update to our software during its life costs it nothing. → Running it without youWhy does my saved schedule never fire?
Because saving it is not the same as switching it on. A schedule can be saved and left off, and a schedule left off does not run. A card works the same way: you approve it first, then start it on its card page or the kill switch page. Switching it on counts as a spending decision, because from that moment the workflow can fire unattended, with nobody present. → Running it without youStrategy cards
What is a strategy card?
A rule in plain English that the agent writes for you in chat. It says where it runs, when it acts, what it does and how much it may spend. A checker refuses any sentence it cannot turn into a safe plan. You read what the card can and can never do, read its backtest, and approve it with your signature. Then you start it. → Strategy cardsWhat is the difference between a card and a workflow?
A card waits and decides by itself, inside limits you signed. A workflow is a plan you run, schedule or chain. The chat writes cards only. It does not write workflows, schedules or chains. Those still exist, and you set them up in the console or through an outside agent. → Running it without youCan I run a card without risking money?
Yes, as a paper run. An approved card acts on live prices with simulated fills, and no money moves. Start it from the approval page and choose “On paper”. The kill switch stops a paper run like any other card. → Backtest and paper runsWhat does my signature cover when I approve a card?
The card, its limits and its budget. It does not cover each trade. A live card spends real money without asking each time, and it can still make a losing trade inside those limits. Approving freezes the version. A change of any kind, even one limit, is a new version that is backtested and approved again. → Going liveHow do I stop a card?
From its card page, or from the kill switch page, where you can stop all your plans at once. Stopping moves no money and reverses nothing. → Going liveDoes a card have a limit by default?
The only default is that one run may move up to $1,000,000. There is no default limit on the steps of a run or on a week’s total. Write the caps you want into the card, and do not rely on a default. A live card needs a budget. → Going liveWhy does my card have no backtest?
Some kinds of card have no model to replay: a resting order, a card that trades several assets, a fee collection, a run-once card, and a Binance card. The approval page says so and lists every order, swap and collect the card will run. You sign that list instead of a report. → What has no backtestWhy was my card refused?
The checker refuses a card it cannot turn into a safe plan. Common reasons are a rule that could fire on every bar with no guard, a price the card does not state, and a swap that accepts too much slippage. The refusal says which line. → Writing a cardHow fast does a card see a price change?
A card wakes when its feed changes. A price older than 15 seconds counts as stale, and a card refuses to act on a stale or missing price. → The feed runtimeYour money
These answers describe what the code enforces.Do you ever hold my funds?
No, and there is no deposit step to skip. Your money sits in an account whose address is worked out from your own address by a fixed calculation. It is the same address on every network, and it exists as an address before anything is deployed to it. There is no vault of ours, no pooled contract, and no address of ours that your funds pass through. You are the root authority on that account and there is nothing above you. Neither the agent nor the key that pays network fees can hold a token at all. That is not the same as the money never moving. “No deposit step” means there is no vault of ours to deposit into, not that your funds stay in the wallet you already use. For a run to spend anything it has to be reachable by your operating address, and you pick how: send tokens to it, or leave them in your wallet and grant an allowance for adeposit.pull step to draw when a run
needs it. Approving moves no money, and either way the address holding it is
one you own outright and can empty without us.
→ Where your money actually sits
How do I get my money out?
You move it yourself, with your own wallet, like any ordinary transfer. The account answers to your signature, so getting your money out is not a request to us. It needs no agent, no fee payer and nobody’s approval. It does not depend on any part of this system being switched on. See also can I get my money out without you, which has two conditions.What exactly may the agent do?
Only what was written into your account when it was set up: named operations, named assets, and a daily limit per token that you set and the agent cannot raise. Two things do that work, and both are built into the design, so nobody can change their mind about them.- The agent cannot write its own instructions. What it signs only supplies the details of an operation. It cannot set “which address to call” or “what to say to it”, so an instruction of that shape cannot be written down.
- Anything that produces value goes to you. The destination is named by what it is, such as your own account, and looked up from your account when the step runs. The plan never gets to name an address.
What if the agent’s signing key is stolen?
The agent is a contract, not a key. Your account gives it permission, and it cannot give that permission to a plain key at all. The key we hold lives inside a hardware enclave and is created there. All it can do is tell the agent contract to make one call. It is not expected to leak. If it did, the key would still reach your account only through that contract. So a stolen signing key cannot take your account. The agent contract refuses, on the network, before the call is made: it will not target your account directly, will not run the instructions that install permissions or change who owns the account, and will not touch an operation the guardian never approved or that you never switched on. It also cannot choose where money goes. Destinations are looked up from your account when the step runs, so the agent cannot set the recipient and there is nowhere to put an attacker’s address. You do not have to wait for us either way. Switching the agent off is the next answer, and your own way out never involved the agent at all.How do I switch the agent off?
One transaction from your own wallet. There are two kinds of switching off.Can I use a protocol you have not integrated?
Not by writing it into a plan. The agent can only use a fixed set of actions. Something is in that set because it was built, deployed, approved for that network and switched on for your account. A plan naming anything else is refused when it is written, rather than attempted and failed halfway. Adding to the set is cheap. A new venue is a new entry and does not change how plans work, so plans written before it existed keep their shape. The fixed set of actions is what keeps you safe, and how quickly we can extend it does not weaken that. → What it can doHow do I know what a plan will do before it runs?
Because you approve the steps themselves. Before anything runs you are shown every step: which protocol it uses, which network it lands on, the amount both as a plain number and as “0.5 ETH”, and where the value goes next. Saving a workflow runs nothing. Running it is a separate decision, taken by a person pressing Start. → What the agent can callWhat runs it
What is actually running my workflow?
A handful of small services of ours, none of which can hold your money. One reads your sentence and writes the plan. One does the waiting, which is most of the work: a network crossing is a few minutes of waiting, and a price you are waiting for might arrive next week. One keeps the live prices that cards read. One is the hardware enclave that holds the agent’s key and authorises steps through the agent contract, and does nothing else. One pays the network fee so you do not have to hold a balance for it. None of them can move your money except through the operations already switched on for your account. → How a run worksWhat happens if your servers restart in the middle of my run?
It picks up where it was, rather than starting again or losing track. The waiting is built so that a service restarted, updated or replaced halfway through a crossing comes back knowing which step it was on. The record of what has already happened lives in a database rather than in the memory of a running process. Nothing is reconstructed by guesswork and no step is silently done twice. One consequence: a workflow waiting for a person is in mid-flight across every update we ship. That is why each step is pinned to the version it was approved against. → How a run worksIs real money moving?
Yes. DeFiLoops is live at app.defiloops.com, on the real Ethereum, Base, Arbitrum and HyperEVM. A run spends real money. The one exception is Binance, which connects to Binance’s demo exchange only. → How a run worksWhat stops a bad transaction reaching the network?
1
The plan is checked against the fixed set of actions
An operation not on that list cannot be named, and there is no way around it.
2
Every amount is written twice
Once as the raw number the machine uses, and once as “0.5 ETH” in plain units.
Machines write amounts as very large integers, so a plan where every amount is a
thousand times too small is consistent with itself and passes every other check.
A misplaced decimal point cannot survive in both versions.
3
Every transaction is run on a copy first
Before it is sent, we run the exact transaction on a copy of the network’s latest state. A run that would do more than the plan allows is refused. This check always runs on Ethereum, Base and Arbitrum. It does not run on HyperEVM.
4
The fee payer is handed the step's details, never a raw instruction
This is the most important check. → What it can do
Why does every operation have a version number on it?
Because a workflow can be waiting for a long time while the software underneath it changes. Without a version pinned to it, a job that outlives an update could start doing something different halfway through, with no warning. A newer version that only adds things satisfies the pin. A newer version that changes behaviour does not, because you approved the old behaviour. When several versions would do, it picks the oldest one that fits. The newest is code your plan was never shown. → What it can doNetworks and trading
Which networks does it run on?
Ethereum, Base, Arbitrum and HyperEVM, with real money. Base has every operation. Ethereum and Arbitrum have almost all of them. HyperEVM is for moving USDC and for trading on Hyperliquid. Your account has the same address on all four. → NetworksWhy can I not swap on HyperEVM?
The swap router prices swaps on Ethereum, Base and Arbitrum only. HyperEVM carries USDC, HYPE and Hyperliquid trading. A plan that names a swap there is refused when you save it. → NetworksHow long does a crossing take, and what does it cost?
USDC usually lands within about two minutes, and WETH within about one. That is an expectation and not a promise. Up to 1% comes off the amount you send, and the plan assumes 0.1% when it checks your balance. The money lands in your account on the far side, never in your wallet. → Moving between networksWhat stops someone sandwiching my swap?
We do not claim a private route. Transactions go out through ordinary network providers. Four things limit the damage: a price floor signed into the step, price checks on that floor, a short expiry of about ten minutes, and a run on a copy of the network before sending. A card or order may set its floor at most 1.5% below the market. At signing, the enclave refuses a floor more than 3% below its own price source, except for a token that source cannot price. → Sending transactionsDoes every transaction get checked before it is sent?
On Ethereum, Base and Arbitrum, yes: we run the exact transaction on a copy of the network’s latest state and refuse it if it would do more than the plan allows. On HyperEVM we chose to send without this check. → Sending transactionsWhy do zero-knowledge accounts behave differently on Ethereum?
The proof costs too much gas there, so on Ethereum a proved account runs without it and works like a standard account. The signature check, the deadline, the destination check, the price floor and the loan ceiling still hold. Your permission list no longer limits the agent. Base, Arbitrum and HyperEVM keep the proof. → Networks and Proved accountsTrading venues
Does DeFiLoops hold my money on Binance, Hyperliquid or Lighter?
No. Binance holds it in your own Binance account. Hyperliquid and Lighter hold it in your own account at that venue. We hold a trade-only key inside the hardware enclave, and it cannot withdraw to an address of its own. → Trading venuesIs Binance live with real money?
No. Binance connects to Binance’s demo exchange only, with play money. Do not send real funds to it. → BinanceCan the agent withdraw my money from a venue?
No. A withdrawal pays your own account. On Binance and Hyperliquid, the venue enforces that on its side. On Lighter, a withdrawal always pays your account on Ethereum. → Trading venuesWhat if your servers go down or are breached?
Your money stays in your own account at the venue. Someone who controls our servers alone can delay or drop requests, but cannot sign an order, because the trading key lives in the enclave. The venue pages have tables of what each kind of breach could and could not do, and how you take control back. → Trading venuesCan the agent open Hyperliquid perpetuals for me?
Yes. On Hyperliquid and Lighter the agent opens and closes perpetuals itself, through a card, inside the account locked to that venue. You connect, deposit and approve the card. An order can carry a take-profit and a stop-loss price. Setting or moving them on a position that is already open is not available yet. → HyperliquidCan I write a card that trades on Lighter?
Yes. A card can name Lighter. Running it needs a verified Lighter connection and USDC on Lighter. → LighterHow do I get out of Lighter without you?
Use the exit kit. It is a page that builds the Ethereum calls you send from your own wallet through your account. It calls none of our services, so it works with every one of our servers down. → LighterWhy does my first Hyperliquid deposit cost 1 USDC?
Hyperliquid charges 1 USDC to open a new account, so send at least 2 USDC the first time. Size your margin from what arrives, not from what you sent. → HyperliquidWhy does a Lighter registration take up to 14 days?
Registering your trading key is a priority transaction on Ethereum. It is usually quick, but Lighter must include it within 14 days. The app checks again by itself. → LighterWhat does revoking a grant do?
It stops our relay from placing orders. It does not undo your signature. A signed grant that has not expired stays valid. The hard stop is replacing the key with the exit kit, or tightening the lock. → Trading venuesSecurity
What are the keys, and what can each one do?
The 3-day delay on approving a new operation is there so a hostile operation is visible
before it can act. It does not cover the upgrader, which can replace the code behind shared contracts at once on every
network today. Each use of the protocol keys needs a touch on a hardware security key. → The contracts
Can the agent drain my account?
No, but it can trade badly. An earlier design put a guard in front of the agent that banned a list of actions. Four attacks got past it on a copy of the real network, because a ban list is always incomplete. All four used a raw call on your account. The agent no longer has any raw call, so the guard was removed, and those four attacks no longer apply. The agent now reaches your account only through a fixed set of operations. Each takes labelled fields, never a target and never an instruction. It can never become the owner or hand your account to somebody else, and it cannot send value to an address of its own. It can still:- move any amount your account holds, because nothing caps size unless you set a daily limit;
- sign a swap or liquidity step priced up to 3% off the market, although our servers hold cards and orders to 1.5%;
- sign a step for a token the enclave’s price source cannot price, with no price check;
- put your money in a position you did not want, inside the approved set.
The daily limit per token is off by default, and the app has no screen for it yet. Today
you set it with a transaction your wallet sends. → Your account and your keys