TL;DR: Monitoring a trading bot well means three things: alerts for the events that need a human (errors, liquidations, safety pauses), a scheduled review of the forward record against the backtest, and pause rules you wrote down before going live. On Trigr, agents send email and web-push alerts per event type, draw a frontier marker between backtest and forward record, and pause themselves when the account or key is not in the state they expect.
Why does a trading bot need monitoring at all?
An automated strategy removes the need to click buttons. It does not remove the need to supervise, which is the part of deploying a strategy as a Hyperliquid agent that starts after launch day. Things go wrong in ways a backtest never shows: a key expires, an exchange rejects an order, someone trades manually on the same account, or the market simply behaves differently from the history the strategy was built on.
The goal of monitoring is not to second-guess every trade. It is to separate two kinds of bad news:
- Broken: the system is not doing what you designed. A missing stop, a rejected order, an unexpected position, a stale connection.
- Unlucky: the system is doing exactly what you designed, and the market is not cooperating. A losing streak within the range the backtest showed.
Broken needs action within minutes. Unlucky usually needs patience and a scheduled review. Most bad decisions around trading bots come from mixing the two up, for example turning off a sound strategy during a normal drawdown, or ignoring a real error because "it is just a bad week."
Which alerts should you turn on?
Alerts are for events that might need a human. Trigr's alert settings let you toggle each event type separately:
| Event | What it tells you | Suggested setting |
|---|---|---|
| Signal fired | A strategy you follow produced a new entry signal | Optional; useful early, noisy later |
| Take-profit / stop-loss | A position hit its TP or SL barrier | On for the first weeks, then optional |
| Agent opened or closed a position | The agent acted | Optional for high-frequency strategies |
| Agent error | Something failed and may need you | Always on |
| Position liquidated | The venue closed a position for margin | Always on |
| Product updates | Important announcements | Your preference |
Alerts arrive by email or web push. Push reaches your device even when Trigr is not open. Trigr is a Progressive Web App, so on Android and desktop Chrome you can install it from the browser prompt; on iOS, web push requires adding the app to your home screen first (iOS 16.4 or later).
A practical rule: route errors and liquidations to the channel you actually see within minutes, which for most people is push, and keep informational events on email or off. If every fill buzzes your phone, you will stop reading the alerts that matter.
Safety emails you cannot switch off
One class of message is not optional. If Trigr detects activity or exposure on a Hyperliquid account that it did not place, such as a manual trade, another bot, or spot activity, it quarantines that exact account. Every agent bound to it shows a persistent Safety paused state, new entries stop, and Trigr always queues a safety email, even if your general agent-error email preference is off. Only ownership-verified managed exits and reconciliation continue until you clear the outside activity and run Recheck and restart.
This is deliberate. An agent that silently adopted a position it did not open would be sizing and exiting against a book it does not understand.
What does the agent already watch for you?
Some monitoring is built into how agents run, so you do not have to rebuild it with alerts. According to the live trading docs:
- Protective orders. Every live entry ships with a reduce-only take-profit and a reduce-only stop-market order. If placing them fails, for example because the venue times out, the agent retries with backoff and marks the trade
protection_failedso it is repaired on the next pass. - Reconciliation every minute. A job compares each agent's records with Hyperliquid's actual state. If the venue no longer reports a position the agent was tracking, the close is attributed (TP, SL, manual or liquidation) from Hyperliquid fill data, and realized PnL comes from those fills rather than from a local price feed.
- Key health. The autonomous trade-only key is valid for about 90 days. Trigr cross-checks Hyperliquid's list of approved agents when it loads an agent; if the key has been overwritten from another device, the agent pauses in a clear "needs reconnect" state.
That key is a Hyperliquid API wallet. It can place and cancel orders but cannot withdraw. Hyperliquid's own documentation on API wallets explains that registering a new unnamed API wallet deregisters the previous one, and that wallets are pruned when they expire. That is exactly why a key approved elsewhere can quietly break an agent, and why the reconnect state exists.
How do you read the forward record?
On an agent's chart, a frontier marker separates the backtested segment from the forward record: live performance for Hyperliquid or Propr agents, or the paper track for paper agents. Everything to the right of the marker happened after you pressed go. That line is the most important thing on the chart, because the left side was chosen with hindsight and the right side was not.
When you review the right side, compare it with what the backtest said to expect, not with what you hoped for:
- Trade frequency. If the backtest averaged a few trades a week and the agent has made none in a month, or dozens in a day, something differs: the data, the market regime, or the configuration.
- Win rate and average trade. Early numbers are noisy. With twenty trades, a strategy with a genuine 55% hit rate can easily show 40% or 70%. Look at whether the result sits inside a plausible range rather than whether it matches exactly.
- Drawdown. Compare the live drawdown with the worst drawdown in the backtest. A live drawdown within that range is uncomfortable but expected.
- Costs. Live fills pay real fees, slippage and funding; Hyperliquid pays funding hourly, while backtest funding uses 8-hour settlements. If your backtest was gross, compare against a net rerun with slippage and funding switched on, or you will mistake known costs for a broken strategy.
Paper agents have their own caveat: they fill at the current Hyperliquid price and deduct the taker fee plus the builder fee, but model no slippage and no funding. A paper record is a check on behavior and signal timing, not a precise preview of live PnL. The guide to paper trading agents covers what a forward test can and cannot prove.
When should you pause a trading agent?
The best time to decide when to pause is before the agent goes live, when you are calm and have the backtest in front of you. Write the rules down. A useful set looks like this:
- Pause immediately for broken behavior. A position the strategy should not hold, repeated errors, a protective order that will not place, a safety pause you cannot explain. Fix first, restart later.
- Pause for a drawdown clearly outside history. Pick a threshold relative to the backtest's worst drawdown, for example well beyond it rather than just touching it. Losses within the tested range are what the strategy was expected to do.
- Pause for a change in the environment. A venue incident, a big change in fees or funding, or a market-structure event the strategy was never tested on.
- Pause for a change in you. If the capital at risk now matters more than you planned, reduce it. That is risk management, not a verdict on the strategy.
Two things are not good reasons to pause: a single bad trade, and a short losing streak inside the tested range. Stopping a strategy after every normal drawdown and restarting it after the recovery locks in the worst part of each cycle.
What pausing actually does on Trigr
A paused agent blocks new exposure while existing managed positions can close or reconcile. It does not market-sell everything. If you need to be flat now, close positions deliberately. When you restart, Trigr reacquires monitoring and rechecks the exact account before entries resume.
Also remember the capital floor: Hyperliquid rejects orders under $10 of notional, so live entries below $11 are raised to $11, and an agent with less than $11 of capital cannot open positions at all. An agent that "stopped trading" after a drawdown may simply have run below that floor.
A simple monitoring routine
You do not need a dashboard habit. A light routine is enough for most people:
- Daily, one minute: scan alerts for errors, liquidations and safety pauses. No action if there are none.
- Weekly, ten minutes: look at the forward record right of the frontier marker. Check trade count, drawdown and net PnL against the backtest's ranges.
- Monthly: re-read your pause rules, check the key's expiry and your remaining margin, and decide whether the capital allocation still fits.
- After any incident: write one line on what happened and what you changed.
If you run several strategies in one agent, review the per-strategy legs as well as the combined curve. A drawdown in one leg can be normal even when the combined curve looks calm, and the reverse.
What this means for you
Trigr's design moves the constant watching onto the system, so your attention goes to the few events that need a person. Keys, protective orders and account state are checked automatically; anything unexpected pauses the affected agents rather than trading through it; and the frontier marker keeps the honest forward record visibly separate from the fitted history. Your job is to choose alerts, set pause rules in advance, and review on a schedule.
Backtests are not guarantees; perps are leveraged and can lose more than expected, and no monitoring setup prevents losses.
Next steps
Before you start a live agent, run through the pre-flight checklist, then set your alert toggles and pause rules while the agent is still on paper. The trading agents docs cover the agent lifecycle in detail, and the strategy marketplace shows each published strategy's forward paper track alongside its verified backtest.