> For the complete documentation index, see [llms.txt](https://doc.lockerprotocol.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://doc.lockerprotocol.com/hyper-agentic-trader/faq.md).

# FAQ: the agent wallet

Short answers about the agent wallet, the keys, the limits and the costs.

### Does the agent ever have my private key?

No. Your account's key stays in your wallet: in the Vault, offline, the safest way; or in your browser extension or mobile wallet, if you chose one of those. The computer holds a second key, the **agent key**, that Hyperliquid lets trade and refuses every withdrawal and every transfer to another account. See [The guardian and the keys](https://doc.lockerprotocol.com/hyper-agentic-trader/guardian-and-keys.md).

### What happens if someone steals my computer?

They get a sealed key file they cannot open without your password, and even with it they cannot withdraw. They could trade your account badly, or deposit it into someone else's Hyperliquid vault, which is why the account should hold only what you are ready to lose. If your account signs with a browser extension on that computer, its own lock is what protects the account key there: one more reason Locker Vault, offline, is the safest. Revoke the key from another machine with `lpa agent revoke`, with your wallet.

### Can the agent withdraw my money?

No. Hyperliquid refuses the agent key every withdrawal and every transfer to another account, and the guardian signs trading actions only. Only your wallet signs a withdrawal, and you read the destination there before you sign.

### Do I need Locker Vault?

It is the safest way, and the one we recommend, but not the only one. The account signs the agent's approval, the mandate and every movement of funds with one of three wallets, chosen once by `lpa setup`:

1. **Locker Vault**, on a phone or tablet kept offline: the key never touches a device that is online, and the Vault shows every request in words.
2. **Your browser extension** (MetaMask, Rabby), on a page `lpa` opens on your computer.
3. **Your mobile wallet** (MetaMask, Rabby, Coinbase Wallet...), by a WalletConnect QR scanned once.

For trading in simulation you need none of them: `lpa setup --paper` needs no wallet and no money. See [Three ways to sign](https://doc.lockerprotocol.com/hyper-agentic-trader/agent.md#three-ways-to-sign-for-your-account).

### Can I sign with MetaMask?

Yes: as your browser extension (`lpa init --with browser`) or as your mobile wallet (`lpa init --with mobile`). An account that already trades on Hyperliquid from MetaMask can be linked as it is. Before any money, one signature that moves nothing proves your wallet signs for that address; a contract address (a smart wallet) is refused. MetaMask shows a typed signature as raw fields, so the local page explains each field first. With a mobile wallet, the requests go through WalletConnect's relay, end-to-end encrypted, and `lpa signer forget` closes the link. `lpa signer use vault` moves the same account to Locker Vault later, keeping the trading key and the mandate.

### What about a hardware wallet?

It is not one of the three ways. Behind your extension or your mobile wallet, `lpa` does not refuse one. Many show only a fingerprint of a typed signature, not its fields, so the page's list is then the only place you read what you sign. Locker Vault shows every request in words.

### Do I need the browser extension?

The Locker Protocol Wallet extension, no. With the Vault, `lpa` reads the same sync QR the extension reads, and the Vault's own page says so. You can use both, on the same account.

### Is simulation (paper trading) realistic?

It fills on Hyperliquid's real order book, pays the real fees (always counting the Locker fee, the worst case), is charged the real hourly funding, and is liquidated like a real cross-margin account. A position's take profit and stop loss fire there as on Hyperliquid, and so does an entry trigger. It also catches up what the price did while nothing ran: with the computer off or the network cut, a stop touched meanwhile fires at the hour of its candle and at its price (at the open when the candle opened past it), a wick that would have liquidated the account liquidates it at its hour, and the hours missed pay the funding Hyperliquid charged then. What it does not simulate is a resting limit order: in simulation, a limit must cross the book or nothing happens. See [lpa paper](https://doc.lockerprotocol.com/hyper-agentic-trader/commands/paper.md).

### What does it cost?

Hyperliquid's own trading fees, a Locker service fee of 0.05 % of each order's notional, 1 USDC to withdraw (Hyperliquid's charge), and the model: the credits of your plan when you use the included one (50 a month on the free plan, about 0.3 a day of the daily pilot), or whatever your own provider charges. A model on your own machine costs nothing. See [Models and costs](https://doc.lockerprotocol.com/hyper-agentic-trader/models-and-costs.md) and [Plans and credits](https://doc.lockerprotocol.com/hyper-agentic-trader/plans-and-credits.md).

### Do I need an API key?

No. Every plan, the free one included, comes with a model, and your sign-in is its key: `lpa hyperkeel login` in the terminal, or the application's sign-in. A key is only for a provider you pay yourself. See [Plans and credits](https://doc.lockerprotocol.com/hyper-agentic-trader/plans-and-credits.md).

### What are credits?

The unit the included model is counted in: a measure of what you asked of it, with a reference of 100 credits for one dollar of the provider's price, not a deposit and not a balance you can withdraw. Each answer says what it cost and what is left; `lpa hyperkeel status` says the plan and the credits left until the end of the month. When they run out, the call is refused with `NO_CREDITS`, the pilot pauses until midnight UTC, and the way out is a bigger plan or a top-up on hyperagentictrader.com, a personal key or a local model.

### How do I pay for a plan?

In USDC on Arbitrum One, from a wallet in your browser (the Locker Protocol Wallet extension or MetaMask), to a locker of Locker Protocol dedicated to Hyper Agentic Trader: the site writes your order, your wallet confirms the exact amount and the deposit, and the plan is active once the chain confirms it, within minutes. Your trading account and the Vault are not involved. See [Plans and credits](https://doc.lockerprotocol.com/hyper-agentic-trader/plans-and-credits.md#paying-for-a-plan).

### Can I use it without sending anything to a model provider?

Yes. Point `lpa config` at LM Studio or Ollama on your own machine, and nothing leaves it. The pilot says what leaves the machine before it starts.

### Which strategy should I use?

The one you have measured. `ladder-long` is in use until you change it, and it is the **measured charter**: the one strategy whose figures two years of candles chose. Every other shipped strategy, and every strategy you write, is said unmeasured until you measure it yourself:

```sh
lpa pilot strategy list
lpa pilot strategy show --name grid-long
lpa pilot calibrate --symbol BTC,ETH --strategy grid-long --starts 6
```

The measure reads the venue's past candles and funding, asks nothing of a model and places nothing. Measure from several starting days: a return measured from one day is a draw, not a figure. Then put it in use (`lpa pilot strategy use --name <name> --preview` first), and give it simulation days before any money. No strategy is a recommendation, here or in the product, and none of them promises a return. See [The strategies](https://doc.lockerprotocol.com/hyper-agentic-trader/strategies.md) and [Measuring the method](https://doc.lockerprotocol.com/hyper-agentic-trader/calibrate.md).

### Why is there no stop loss?

Because the strategy in use by default builds the position in steps instead, each one improving the average, and exits the whole position on one limit at the average plus a small target. That is a deliberate, measured choice with a ceiling on the engagement. If you prefer a stop, the kinds `stop` and `range` place one, and `lpa pilot strategy use` switches with one command. Read [How it trades](https://doc.lockerprotocol.com/hyper-agentic-trader/how-it-trades.md) before you choose.

### Can the pilot run without a model at all?

Yes, with a strategy the code decides. `donchian` ships that way: it arms a breakout of the last thirty days each day, with a trailing stop, and asks no model, reads no brain and spends nothing. `lpa pilot run-now`, `lpa pilot replay` and `lpa pilot schedule --on` then need no model set. Under every other shipped strategy a model decides: the side and the market, the entry by steps or with a stop, the borders of a grid or a range when the measured ones are wrong.

### Can I have the pilot change its method by itself?

Only among the strategies you allowed. `lpa pilot strategy choice --among <two to five>` lets it pick one each day, by a table of the regime of the leading market or by the model in one brief call that can only answer with a name on your list; `--today` says what today would choose and writes nothing. A day that cannot choose plays the strategy in use, and says so. Nothing else changes a method on its own, and no strategy widens a limit: `lpa pilot strategy show` says what the policy and the mandate would refuse of one before it ever plays.

### Can the pilot trade my real account?

Yes, and only on a gesture of yours. The continuous pilot trades for real with `lpa pilot start --live`, under a mandate whose uses include `pilot`. The **daily** pilot plays on your real account with `lpa pilot schedule --on --live` (one day every day, by the service) or one day with `lpa pilot run-now --live`; both start only with the account of this computer, a key in the guardian and a mandate in force that allows `trade`. Without `--live`, both stay in simulation.

### Can an AI agent raise my limits?

No. It can edit `policy.json`, and gain nothing: the guardian keeps the policy it was unlocked with, so a wider limit waits for **your** password. And nothing on the computer widens the mandate, which only your wallet signs.

### Can an AI agent place an order on its own?

Not without your go. Without a terminal and without `--yes`, an order answers `CONFIRMATION_REQUIRED`. Through the MCP server it needs a `quote_id` you saw, good for ten minutes and once. The exception you choose yourself is a mandate that allows `pilot`: then the pilot opens positions without asking, within every limit you signed.

An agent that only wants to know whether an order would pass has a command for it, which signs nothing and writes nothing, not even a refusal in the journal:

```sh
lpa perps check --symbol BTC --side long --size 0.01 --leverage 3
```

```
LONG 0.01 BTC at 3x (market)
  entry        83205   (mid 83199.5, worst accepted 85695)
  notional     $832.05   margin $277.35
  fees         0.095% ($0.7904)   (base tier)
  liquidation  56172.152   (estimate, this order alone, cross margin)
Would be refused: The policy refuses this order: $832.05 is above the $100 per order.
```

### How do I stop everything right now?

```sh
lpa panic
```

Every pilot and every copy stopped, then the key out of memory. It closes nothing: `lpa perps close --all` does that. On the live page, type `STOP`.

### How do I get my money out?

Close the positions, then withdraw: `lpa perps close --all`, then `lpa perps withdraw --amount 25`, which sends the USDC to your own address on Arbitrum One, or to an address you name with `--destination`. Your wallet signs the withdrawal, as it signs every movement of funds. [Stop, and take your money out](https://doc.lockerprotocol.com/hyper-agentic-trader/leaving.md) walks the whole way, stopping the pilot, retiring the trading key and removing the program included.

### Why does it ask for my password so often?

Because the password is what hands the key to the guardian, and a guardian holds it for eight hours by default. `lpa unlock --hours 24` is the longest a terminal allows. A service holds it for the agent key's whole life, until you lock it or the machine restarts.

### The mandate expired and the pilot stopped. Is that a bug?

It ended with the agent key, or on the shorter term you signed: a mandate lasts the key's own life by default (180 days at most) and never outlives it. `lpa agent renew` then `lpa mandate sign` sign a new key and a new mandate; the renewal is the moment you read the limits again.

### I renewed the agent and nothing opens any more.

A mandate signed for the previous agent opens nothing for the new one, and the command says so. Sign a new mandate.

### I installed `lpa` on a second computer (or a server) for the same account, and the first says the agent is not listed any more. Why?

Because Hyperliquid keeps **one agent per name per account**. `lpa setup` on the second machine made its own trading key and had your wallet approve it under the same name, which replaced the first machine's key. Nothing was stolen and no money moved: the first machine simply has a key Hyperliquid no longer lists, and it says so, in `lpa status` and in `lpa doctor`:

```
FAIL  agent        0x8c1f...9a2e, not listed by Hyperliquid: lpa agent renew. Another lpa under the same name (a computer or a server) replaced it, or the account was emptied.
```

`lpa agent renew` on the first machine approves a new key there, with your wallet, and replaces the second one's in its turn. Two machines on one account therefore take turns, and the one that renewed last is the one that trades. To run both at once, give each machine its own account: one recovery phrase can hold several, and `lpa init --account 1` picks another from the Vault's sync QR.

### What is hyperkeel, and do I need it?

Our market data service for Hyperliquid: the brief, the leaderboards, followed traders and alerts. It is optional, nothing else needs it, and it never receives your wallet address.

### Can I run it on a server?

Yes. `lpa service install` runs the guardian round the clock with a small console on `127.0.0.1:7717`, which you reach over your own SSH connection. The signing pages go through that console, so the Vault (or your browser extension) is still on your desk; a mobile wallet is reached by the server through WalletConnect's relay, its QR printed in the terminal. See [Install on a server](https://doc.lockerprotocol.com/hyper-agentic-trader/install-on-a-server.md).

### Does it work on Windows?

Yes, with the PowerShell installer. There is no system service on Windows: `lpa agentd --service` in a terminal left open does the same job.

### Does it send any telemetry?

None. The hosts it talks to are listed in [Files and folders](https://doc.lockerprotocol.com/hyper-agentic-trader/files-and-record.md): Hyperliquid, Arbitrum nodes, hyperkeel when you use it, your model provider if it is not local, a few public sources for the pilot's dossier, WalletConnect's relay only when your account signs with a mobile wallet, whatever a plugin you installed reaches, and `127.0.0.1`. No analytics, no crash reports, no update check.

### Where do I see what it did?

`lpa journal` for what was done, `lpa pilot day` for a day of the daily pilot, `lpa pilot plans` for what it holds right now, `lpa store tokens` for what the models cost, and `lpa live` for everything as it happens.

### Do I need the application, or the command line?

Either. The application is the product: it installs the engine, signs you in, and gives you the workshop, the assistant, the pilot and your portfolio on screens, with the model included in your plan. The command line, `lpa`, is the same engine for a terminal, a server or a script, and this guide's command reference is its words. The MCP server is the third face, for an agent in another program. All three share one folder and the same refusals. See [What Hyper Agentic Trader is](https://doc.lockerprotocol.com/hyper-agentic-trader/agent.md).

### Where do I get help?

[Contact and community](https://doc.lockerprotocol.com/help/contact.md).

## Learn more

* [What Hyper Agentic Trader is](https://doc.lockerprotocol.com/hyper-agentic-trader/agent.md)
* [The application](https://doc.lockerprotocol.com/hyper-agentic-trader/application.md) and [Plans and credits](https://doc.lockerprotocol.com/hyper-agentic-trader/plans-and-credits.md)
* [The strategies](https://doc.lockerprotocol.com/hyper-agentic-trader/strategies.md)
* [Safety rules for the agent wallet](https://doc.lockerprotocol.com/hyper-agentic-trader/safety-rules.md)
* [Command reference](https://doc.lockerprotocol.com/hyper-agentic-trader/commands.md)
