TL;DR: A Hyperliquid trading bot on Trigr is an agent: a venue plus up to 52 strategies, one per market, each with its own allocation and risk rules. The path is backtest honestly, forward-test on a paper agent, then connect Hyperliquid with a trade-only agent key that can place orders but cannot withdraw or send funds to another address, and go live with sizing above the $11-per-order floor. You keep your master wallet and your funds throughout.
What does "deploying a strategy" actually mean?
Deploying means taking a rule set you have tested on history and letting software act on it in real time, without you clicking buttons. On Trigr the unit of deployment is an agent. An agent has one venue (Paper, Hyperliquid, or Propr prop-firm challenge accounts), one capital figure, and up to 52 strategy slots.
Each slot holds one strategy on one asset. Allocation selects a share of current equity; effective strategy size determines target margin inside that share. Allocation does not reserve cash, and totals can exceed 100%. An entry whose margin does not fit is skipped; a fee-only shortage can reduce its size. Once the agent is active, it listens for new signals from its strategies and opens and manages trades on its own.
The important word is "tested." A bot that trades a strategy nobody has stress-tested is just a faster way to lose money. So the deployment path has three stages, and skipping one usually shows up later as a surprise.
| Stage | What you learn | What can still fool you |
|---|---|---|
| Backtest | Whether the idea had an edge on history, net of costs you choose to include | Overfitting, look-ahead bias, optimistic fills |
| Paper agent | Whether signals fire as expected in real time, on live prices | No slippage or funding in paper fills |
| Live agent | Real fills, real fees, real funding, real margin | Small samples; regime changes |
Step 1: How do you know the backtest is worth deploying?
Start with a backtest you can trust. Trigr's engine uses point-in-time data, so every feature only uses information available at the bar's close, and fills happen at the next bar's open after a signal. If a take-profit and a stop-loss are both touched inside one bar, the stop wins. Those choices make results more conservative than many retail testers; the reasoning is covered in point-in-time backtesting.
Costs matter as much as signals. Trading fees and the builder fee are always applied, but slippage and funding are opt-in and off by default, so a first result is labeled gross. Before deploying, rerun with a slippage assumption (flat basis points per fill) and historical funding switched on. Pick values you would defend: a few basis points per side for liquid majors is a common starting point, more for thinner markets.
Then read the result properly:
- Deflated Sharpe Ratio (DSR). An ML optimization run reports DSR, which adjusts the Sharpe ratio for the number of trials and non-normal returns (Bailey and López de Prado's method); a standard backtest marks it as not yet computed. A high raw Sharpe with a weak DSR is a warning.
- Probability of Backtest Overfitting (PBO) is reported as a diagnostic by the same ML optimization run.
- The trade log and audit trail, which flag which inputs were real and which were simulated.
- The out-of-sample split holds out the last 30% of the full available history by default, reported beside the in-sample 70%. It is unseen only until you tune or pick on it.
If you are unsure how to weigh these, how to read a backtest walks through each panel. A strategy that only looks good gross, or only on one short stretch of history, is not ready.
Step 2: Why forward-test on a paper agent first?
A paper agent runs the same signal pipeline as a live agent but fills orders in a virtual portfolio at the current Hyperliquid price. It answers questions a backtest cannot: does the strategy fire as often as expected, do signals arrive on the bar you expect, and does the combined configuration of slots behave sensibly?
On the agent's chart, a frontier marker separates the backtested segment from the forward record. Everything to the right of the marker happened after you pressed start, so it cannot have been fitted. Paper fills deduct a base-tier trading fee and the builder fee but include no slippage or funding, so treat paper results as a check on behavior, not as a forecast of net returns.
Paper agents are also where TradFi markets live for now. Gold, oil, the S&P 500 and single stocks listed on Hyperliquid through HIP-3 can be backtested and run on paper agents, but live execution on TradFi markets is not available in the current beta. If an agent includes a TradFi slot, it must be a paper agent.
How long to paper trade depends on trade frequency. A strategy that trades several times a day produces a meaningful sample in weeks; a weekly swing strategy may need months. Our article on paper trading a crypto bot covers what to compare and when to stop.
Step 3: How do you connect Hyperliquid without giving up custody?
Hyperliquid has a native feature for exactly this: API wallets, also called agent wallets. You sign one message with your master wallet approving a separate key to trade on your account. According to Hyperliquid's exchange endpoint documentation, approving an agent, withdrawing and sending USDC are user-signed actions, which the master wallet must sign. The agent key can place and cancel orders, not withdraw.
On Trigr, an autonomous agent must keep trading while your browser is closed. So Trigr generates a separate trade-only key for your wallet and stores it encrypted server-side. That key:
- places and cancels orders on the account it is assigned to,
- cannot withdraw funds or send them to another address,
- can also sign two account-management actions for you, creating a subaccount and moving USDC between your own main account and subaccounts, so funds only move between accounts you own,
- is valid for about 90 days, after which you re-approve it,
- can be revoked from your master wallet at any time.
You keep the master wallet key and custody of the funds. Trigr does hold the trade-only key for autonomous agents, and it is worth being precise about that: a trade-only key can still open bad trades, which is why the agent's own risk rules and Trigr's safety checks matter. The agent wallet deep dive covers the model in detail, and the Connect Hyperliquid docs list every signature in the flow. During the beta, live agents also sit behind a one-time access password entered on the Trading Agents page.
Funding the account
Live agents trade collateral already in your Hyperliquid account. Hyperliquid will not credit a deposit below 5 USDC. Trigr's funding modal can bridge tokens from other chains or buy USDC with a card, and it always deposits into your own Hyperliquid main wallet.
Main account or subaccount?
By default, an agent trades your main Hyperliquid account, with no volume requirement. Hyperliquid only allows subaccounts after $100,000 of trading volume, per its sub-accounts documentation. Once you qualify, a dedicated subaccount gives stronger isolation.
That isolation matters because the agent's capital figure is a sizing and accounting limit, not a wallet boundary. Positions on different markets in the same account share margin and account risk.
Step 4: How should you size a live agent?
Sizing is where many first deployments fail quietly. Three numbers matter:
- Minimum order size. Runtime can raise an undersized entry to the rounded venue minimum only when it fits the allocation budget and other checks. Builder warnings allow saving, including when facts are unavailable. A saved configuration does not guarantee an entry. Start and funding guards still apply, and a flat agent pauses when its available balance falls below the floor plus a small fee cushion.
- Allocation times size. Size is the share of the slot's slice committed as margin, and leverage multiplies it into notional. A slot with a 20% allocation and a 10% position size commits 2% of agent capital as margin per entry; at 3× leverage that is 6% of agent capital in notional exposure. On Hyperliquid, the agent sizes from the account's total marked equity, read fresh before entries.
- Concurrency. Max concurrent positions per strategy and agent-wide determine peak exposure. The setup wizard's risk panel shows total exposure and flags over-leverage before you go live.
A practical rule is to go live at a fraction of your intended size, check that fills, fees and P&L attribution match what you expected, and scale up only after that. For how size, leverage, ATR trails and time stops interact, see crypto position sizing and risk settings.
Step 5: What does the agent do once it is live?
A live agent is not a fire-and-forget script. On each cycle it reconciles its own records with Hyperliquid's actual account state:
- Each entry order ships with reduce-only protection: a take-profit order and a stop-market stop-loss at the levels the strategy defines. If placement fails, for example because Hyperliquid times out, execution records the error and retries about once a minute until the orders rest on Hyperliquid. Until then the position stays open without them, and no alert is sent.
- A reconciliation job checks each agent against Hyperliquid every 10 minutes while it has managed positions, and hourly when idle. When a position disappears, the close is attributed (take-profit, stop-loss, manual, liquidation or reconciliation) from Hyperliquid fill data.
- Realized P&L comes from Hyperliquid fills, not from Trigr's own price feed.
- Pausing blocks new entries while existing managed positions can still close or reconcile.
Keep the account exclusive
Assigning an agent reserves that exact account for Trigr execution, including while the agent is paused or flat. If Trigr detects external activity, such as a manual trade or another bot, it safety-pauses every agent bound to that account and emails you; only verified managed exits and reconciliation continue. Before trading that account by hand, pause and settle the agents and move or delete them. Only a dedicated subaccount removes the chance of a simultaneous manual order racing the agent.
What does running a live agent cost?
There are no spreads or markups on your orders. Per trade you pay Hyperliquid's own fees plus a Trigr builder fee of up to 0.05%, which Hyperliquid collects through its builder code mechanism after you approve it once.
| Cost | Who charges it | Notes |
|---|---|---|
| Taker / maker fee | Hyperliquid | 0.045% / 0.015% at the base tier; lower with volume |
| Builder fee | Trigr, collected by Hyperliquid | Up to 0.05% per trade |
| Funding | Paid between longs and shorts | Hourly on Hyperliquid; can be received or paid |
| Subscription | Trigr | Optional; live trading works on every plan, including Free |
Live trading is not gated to a paid plan. Plans change credits for backtests and ML runs, saved-strategy limits and paper-agent counts; details are on the pricing page.
Where do AI assistants fit in?
If you build strategies with ChatGPT, Claude or Codex through Trigr's MCP server, the assistant can research, backtest, save drafts and even create a paper agent, which starts paused. It cannot start execution, place a live trade, withdraw, or touch exchange credentials. That boundary is deliberate: the step from paper to real money stays with you. The assistant works under manual or autonomous approval modes, and either way live execution is out of its reach.
What should you check before and after going live?
Common first-week surprises
- The agent sits paused or trades larger than expected. Usually the $11 floor: an agent whose capital is below it opens no new positions, and undersized entries are raised to $11 of notional. Add collateral or raise the allocation.
- A strategy skipped an entry. When allocations add up to more than 100%, an entry that does not fit the free margin left is skipped. The strategy's row shows it as a skipped entry for insufficient free margin.
- Fewer trades than the backtest implied. Check the timeframe. A 4H strategy that averaged two trades a week in a backtest can easily go a week without a signal.
- P&L below the paper record. Paper fills include a base-tier fee but no slippage or funding; live trading includes all three. Compare live results against a backtest run with slippage and funding switched on, not against the gross one.
- A "needs reconnect" state. The agent key expired or was replaced from another device. Re-approve it from your master wallet.
A deployment checklist
- The backtest is positive net of slippage and funding, and the DSR from an ML optimization run holds up.
- The paper agent has run long enough to produce a meaningful number of trades, and its signals match the backtest's behavior.
- Hyperliquid is funded, the agent key is approved, and you know how to revoke it.
- Every entry clears the $11 notional floor at your chosen allocation, size and leverage without being raised.
- The risk panel shows total exposure and leverage you would accept on a bad day.
- You will not trade the same account by hand, or you use a dedicated subaccount.
Backtests are not guarantees, and perpetual futures are leveraged instruments that can lose more than you expect. Start small and treat the first weeks live as another test.
Next steps
Read the trading agents documentation for the full setup flow, from venue choice to the combined backtest and risk panel.