Over the first weekend of November, X1 founder Jack Levin built a trading system that copies a crowd instead of a person. On November 4 he shared it on X with a single line, "Something I've built over the weekend", and a link to a 12-section paper titled Intelligence Consensus Architecture: Mood Ring, Copybot & Pool Snapshots. In the X1 World Telegram he posted the live dashboard behind it, the Mood Ring sentiment page.
The system now has a public front door. X1 Terminal, the Hyperliquid perp DEX Levin launched on October 30, has a new Auto Trader section that lets anyone point the bot at their own Hyperliquid account. Levin says the fees users pay will go to buying back XNT.
The idea: copy a measurement, not a leader
Copy trading on Bitget, Bybit, Binance, OKX and eToro works the same way everywhere: you pick one trader and mirror them. You inherit their edge, and also their bad weeks, their tilt and the day they change strategy or leave. Leaderboards make it worse, because the trader at the top this month is partly there by luck, and followers tend to arrive just as that luck runs out.
Levin's paper turns this around. The system follows "the net, dollar-weighted positioning and order flow of a pool of 37 independently profitable traders." Individual mistakes partly cancel out, and what survives is the conviction the group shares. The paper compares it to ensemble methods in machine learning and to Numerai, the hedge fund that combines thousands of anonymous models into one.
The 37 Hyperliquid accounts were hand-picked for months of steady profit, active day or swing trading, and different styles: scalpers on majors, traders of stock perps such as NVDA, GOOGL and MU on the xyz markets, Brent and gold traders, pair traders and multi-day position holders. Seven are large, consistently profitable Zcash traders. Each gets a nickname like "Chip Whale" or "HYPE Monk". Together the pool shows +$12.2M over 7 days, +$36.5M over 30 days and +$347.8M all-time, according to the paper.
How the Mood Ring scores a market
A listener follows all 37 accounts over WebSocket and records every fill and position. For each market (BTC, ETH, SOL, HYPE, ZEC, NEAR, XRP and the xyz markets GOOGL, NVDA, SOXL, MU, INTC, AAPL, SNDK, SKHX, XYZ100, gold and Brent) it builds a score from -100 (everyone short) to +100 (everyone long). Three rules shape it:
- No whale can dominate. Any one trader's contribution is capped at 40% of the total in that market. A trader above the cap is flagged "dominant" on the dashboard.
- Positions and fresh trades both count. 60% of the score comes from what the traders hold, 40% from what they bought or sold in the last 24 hours, so it reacts to new conviction without jumping on noise.
- Weak agreement counts as none. Scores between -15 and +15 are treated as flat and the bot holds no position.
Labels run from strong bearish (-60 or lower) through neutral to strong bullish (+60 or higher). On the morning of November 7 the Mood Ring was firmly short crypto: BTC at -52, with 12 of the 14 traders holding bitcoin on the short side ($125.6M short against $9.2M long), ETH at -53 and SOL at -39. The pool was long a few stocks, with the xyz market SKHX at +86 (all seven holders long) and AAPL at +43. The cap was visibly at work. One Zcash whale held 76% of the pool's gross bitcoin exposure and was cut back to 40% in the score.
The copybot
The copybot turns those scores into positions. A score of -52 becomes a weight of -0.52, and the target position is that weight times the follower's equity times their leverage. Roughly every five minutes it compares what the account holds against the targets and sends the difference as opens, adds, reductions or closes.
It trades as a maker first. Orders go in as post-only limits at the best bid or ask, and only if they sit unfilled for about 60 seconds does the bot cancel and finish the order at market. Full closes go straight to market. On the day the paper was written, about 9 of 12 opens filled as maker. That saves money: the paper measured an all-in cost of about 6.0 basis points per maker fill against 8.5 for a market order, both including the 0.05% builder fee. At $100,000 of daily volume that is roughly $560 a month back in the trader's pocket.
Snapshots: a 48-hour flight recorder
The third piece is a snapshot bot that runs just before each copybot cycle. Every five minutes it saves the entire pool to a database: every trader's positions, the scores, the rankings, the bot's own equity, positions and stops, the builder fee rewards, and every fill. It keeps exactly 48 hours, about 576 snapshots, and prunes anything older on each write.
The point is accountability. Anyone running the system can ask what the crowd held at any moment, list changes for one coin or one trader over a time range, or diff two cycles to see exactly which positions, signals and stops moved. When the bot makes a trade, there is a record of the exact crowd state that caused it.
Auto Trader: how it works in X1 Terminal
Auto Trader puts all of this behind a few buttons on the Terminal's API page. Setup runs in this order:
- Connect your wallet. Rabby, X1 Wallet or Phantom. This is the owner wallet, and its key never leaves it.
- Approve the 0.05% X1 builder fee. Hyperliquid requires every user to sign this approval for a builder. The bot refuses to trade without it.
- Create a dedicated trading key. The Terminal generates a fresh Hyperliquid API wallet named "X1 AutoTrader", valid for 90 days, and your owner wallet approves it on Hyperliquid. This agent wallet can place and cancel orders but cannot withdraw funds.
- Seal and register it. The browser encrypts the agent's private key to the execution service's public key (a NaCl sealed box) and wipes it from the page. That lets the bot trade while the tab is closed. Registration always starts paused.
- Choose settings and start. Leverage from 1x to 5x (default 3x), which markets to trade, a minimum order (default and floor $10) and a daily loss stop (default 10%, up to 50%). The account needs at least $15 of equity.
From there the executor does the work. Every five minutes it reads the Mood Ring, checks that the signal is genuine and less than ten minutes old (stale signals abort the cycle), sizes each market and trades. If equity falls below the day's starting value by the loss-stop percentage, it stops opening and adding but can still reduce and close. A table on the page shows each cycle's decisions market by market: the score, the target exposure, the current position and the reason for every action or non-action.
The controls stay with the owner. Every request to save settings, start, pause, close or remove is signed by the owner wallet with a one-time nonce and timestamp, so nobody else can change or switch off the bot. Pause stops future trading. Close all pauses the bot and immediately sends reduce-only market closes with at most 1% slippage. Remove deletes the registration and the encrypted key from the service.
One thing to know before switching it on: Auto Trader manages the whole trading account it is pointed at, including positions you opened by hand in the same markets. That is why the page also supports Hyperliquid subaccounts, which keep a separate balance for the bot (Hyperliquid unlocks them after $100,000 of trading volume). The page is direct about the rest too: "Loss controls do not guarantee a maximum loss. No returns are promised."
How agents run it
"Agent" here has two meanings, and both are built in. The first is the Hyperliquid API agent wallet above: a key with trading rights and no withdrawal rights, which the execution service uses to trade on your behalf. Your funds never move to it. The page says plainly not to send money to the agent address, only to the trading account.
The second is your own software. The same page generates a key file for scripts and AI agents, with a command-line tool that checks access, previews an order from a JSON file and only submits it when given an explicit --execute flag. A developer can give a bot or an AI agent trading access to a Hyperliquid account through X1 Terminal without ever handing over the wallet that holds the money, and every order it places pays the same builder fee.
Why it stands out in the perp DEX world
Hyperliquid has plenty of wallet trackers. HyperIndex, HyperStats, WhaleRoom and CoinGlass will show you what smart money holds, but, as Levin's paper puts it, they "mostly stop at information" and leave the trading to you. HyperDash offers copy trading of up to three wallets. On centralized exchanges, copy trading is custodial and leaders usually take 10% to 30% of followers' profits.
Auto Trader combines the parts those products split up: a consensus signal built from 37 traders with a cap on any single whale, execution that runs on its own, keys that cannot withdraw, a 48-hour audit trail, and a flat fee per trade instead of a cut of profits. Levin's revenue grows with volume routed through the Terminal, not with how much a follower wins or loses, which the paper describes as aligning incentives. That volume is growing fast. The paper, dated November 3, put the Terminal's builder rewards at about $59. By the morning of November 7, Hyperliquid's own API showed $1,392.53 for the X1 Terminal builder address, more than 23 times as much in four days, across all trades routed through the Terminal.
The paper is open about the weak spots. A fixed, hand-picked list can decay, and there is no automatic re-ranking yet. If many followers trade the same signal, their own orders move the price. Traders can converge in strong trends, reducing diversification exactly when it matters, and the xyz stock and commodity markets are thinner than the majors. Numerai, the paper's own example, lost 17.4% in 2023: consensus reduces risk, it does not remove it. The roadmap answers several of these: automatic rotation of the weakest members, performance weighting within the 40% cap, a monitor that cuts weights when traders converge, stops and trailing stops for all followers (today the 8% exchange-side stop and the trailing stop run on Levin's own "Blue Bear" account), and a public read-only snapshot API.
Fees buy back XNT
When Levin launched the Terminal he said the fees collected "will be used to buy and rewards users with XNT", turning trading on Hyperliquid into buy pressure for X1's native token. He has not yet published the details of how the buyback will work.
That is the link to X1. A trader who never touches an X1 app can still follow 37 profitable Hyperliquid traders through Auto Trader, keep custody of their funds, and the fees from that trading go to buying back XNT.
Read the paper: extended whitepaper · short version. Watch the signal: Mood Ring. Set it up: X1 Terminal Auto Trader. Not investment advice.