Trading agents

Automate your strategy with an agent

A trading agent runs your strategies for you. An agent is a venue plus up to 52 strategies, each with its own capital allocation and risk rules. Once active, it listens for new signals from those strategies and opens trades on its own: on paper, on Hyperliquid, or on Propr.

Anatomy of an agent

Venue: paper, Hyperliquid (HL), or Propr. Paper agents are fully simulated; HL and Propr agents place real orders.
Capital: one capital figure for the whole agent.
Up to 52 strategies, each in a slot with an allocation %: the share of equity that strategy may use (up to 100%). Each entry puts up equity × allocation × the strategy's own RISK size as margin: a 20% allocation on a strategy whose RISK size is 50% uses 10% of equity per entry, and the position is that margin times the strategy's leverage. Allocations may add up to more than 100% to use capital more fully: total open margin never exceeds equity, so an entry that does not fit the free margin when it fires is skipped, first come first served, and when several strategies signal on the same bar the one higher in the list enters first. On Propr, equity is the challenge's starting balance, and the challenge's daily-loss and drawdown rules still apply to the whole account. On Hyperliquid and Propr, a strategy whose order sizes under the $11 minimum is raised to it when its allocation covers the minimum, and otherwise waits until the account grows; the builder warns about it but never blocks the deploy.
Distinct-asset rule: the strategies must trade different assets. The UI blocks duplicates and the server rejects them, so an agent is always genuinely diversified across markets.

The frontier curve

On an agent's chart, a frontier marker shows where the agent went live: the transition from the backtested segment to the forward record (the canonical paper track, or live performance for HL / Propr agents). Everything to the right of the marker is real forward-tested performance, not a fitted curve.

Each strategy also keeps a canonical paper track: one system-run forward paper record that starts when the strategy is set running, sized at a fixed slice of its book so the curve is comparable across strategies. This is the live-perf number the marketplace shows. It follows the backtest engine's rules: every RISK onOpposite policy (reverse, ignore, reduce_only), the ATR trailing stop, and the maxBarstime stop, checked in the engine's order (trailing stop, then TP/SL, then the time stop). After a time stop or an intrabar trailing stop, the same bar's entry signal can open the next trade at the next open, as in a backtest. A strategy with a backtest-only node (see backtest-only vs forward-runnable) has no paper track.

Creation flow

Hit + Create agent from the Trading Agents page and the wizard walks you through:

Venue: paper, Hyperliquid, or Propr. Paper agents run against virtual capital; HL/Propr place real orders.
Strategies & allocation: add up to 52 strategies (yours or ones you subscribe to in the marketplace), each trading a distinct asset, and give each one an allocation (the share of equity it may use; see above). Equal, By Sharpe and Inverse vol set the allocations for you, and “Fill every asset” adds the best strategy for each asset.
Combined backtest: preview the agent as a portfolio, with each strategy's returns scaled by its allocation and combined into one equity curve, plus a per-leg breakdown. It costs 50 credits a run, like a Studio backtest, and the same mix is free again for the rest of the UTC day. The builder and Discover cart run it automatically after strategy changes settle for 3 seconds. Allocation changes show a provisional result where possible. Press Run exact backtest to confirm new allocations for 50 credits. An exact mix already paid for that UTC day replays automatically for free. If credits run out, add credits and press Retry to resume. The Propr rules check and community publication keep their explicit paid-run controls. A one-strategy Agent and Studio share one daily charge for the same graph and settings. Changing capital or allocation does not add another charge for that strategy.
Risk panel: review total exposure, max concurrent positions (per strategy and agent-wide), and per-asset exposure; over-leverage is flagged before you go live.

Paper vs live

Paper

Paper agents simulate fills against the current HL price and accrue P&L in a per-agent virtual portfolio. Each fill pays the Hyperliquid base taker fee (4.5 bps) and the builder fee, and liquidation is modelled; slippage and funding are not. (The marketplace paper track charges fees and realised funding, without slippage.) Use paper agents to validate a configuration before risking real capital.

Live (Hyperliquid & Propr)

Live agents place real orders on a venue you control: Hyperliquid or Propr. Live trading isn't gated to a paid plan: anyone who connects a venue can run a live agent. The capital field is a sizing and accounting limit, not wallet isolation; use a dedicated Hyperliquid subaccount when eligible if you need account-level isolation.

Non-custodial HL connect

Hyperliquid agents trade through a non-custodial agent wallet. You approve the agent key once with a single signature; from then on Trigr can place and cancel orders on your behalf without further prompts. The agent key is trade-only: it cannot withdraw your funds (a withdrawal needs your master wallet signature), and you can revoke it at any time. Approvals expire and prompt a quick re-approval. See Connect Hyperliquid for the full flow.

Keep manual trading separate

The default Agent path trades your main Hyperliquid account and has no volume requirement. Compatible Trigr Agents may share the exact account on different markets; same-market conflicts are blocked. Assigning an Agent reserves its selected account for Trigr-only execution, including while it is paused or flat. Before manual trading or using a non-Trigr bot, pause and settle every bound Agent, then move or delete every Agent assigned to that account. Disconnecting the key alone does not release an assignment. A connected key with no assigned or unsettled automation does not reserve the main account. Adding collateral is allowed; do not withdraw or transfer it out while managed positions are open. Proven external activity or exposure, including spot activity, quarantines the exact account, shows a persistent in-app Safety paused state, blocks every bound Agent's new entries, and queues a mandatory safety email. Only ownership-verified managed exits and reconciliation continue. While an Agent is running or its book is reconciling, Trigr actively monitors the exact account. A clean paused and flat assignment may release continuous polling capacity; Recheck & restart always reacquires monitoring and performs a deep, fail-closed account check before entries resume. Clear external positions and orders before restarting. Hyperliquid subaccounts require at least $100,000 of owner volume. Because Trigr cannot lock out a simultaneous wallet order, only a dedicated Trigr-only subaccount removes that venue-level race.

Agent lifecycle

Active: listening for signals. Every new signal that matches the agent's strategy list and available margin opens a trade.
Paused: blocks new exposure while existing managed positions can close or reconcile. A clean paused and flat assignment may release continuous polling capacity; restart reacquires monitoring and rechecks the exact account before entries resume.

How the agent manages positions

On each tick the agent reconciles its DB with HL's actual state: tracked position state and TP/SL orders are re-checked against the venue. Unmanaged external exposure is never silently adopted as a Trigr trade; it blocks new entries and safety-pauses the account's Agents. Closed trades are attributed to a reason (TP / SL / manual / liquidation / reconciliation) and realised PnL comes from HL fills, not from our local price feed.

Trailing stops and time stops

A strategy's ATR trailing stop (trailAtr) and maxBars time stop run on Paper, Hyperliquid, and Propr Agents:

Anchored on the entry fill. Bars that closed before the entry filled are never used to judge the new position.
Time stop. Counts the bars that exist in the market data since entry (a TradFi weekend or session break adds none) and closes the position at market when maxBars is reached.
Trailing stop. On Paper and Hyperliquid the protective stop ratchets after each closed bar and only ever tightens; it never loosens. A bar that closes through the trailing level closes the position at market.
Propr. The time stop and the bar-close trailing cross run as market closes. Moving the protective stop at Propr is off for now, so a Propr position keeps the stop it was opened with.
Re-entry.After a time stop, or a trailing stop the bar crossed intrabar, a Paper Agent can re-enter off that same bar's signal, as the backtest does. Hyperliquid and Propr Agents close only; the next entry comes from a later bar's signal.