What a bot is, honestly
A bot is a set of rules that trades without you. You write the rules as a strategy script, point it at a market and a timeframe, and it acts every time a candle closes.
Two things worth saying before you build one.
A bot removes emotion, not risk. Its whole advantage is that it does not hesitate, does not revenge trade, and does not talk itself out of a stop. It is still trading whatever rules you gave it, and a bad strategy executed perfectly loses money faster than a bad strategy executed nervously.
A backtest is not a promise. It is very easy to tune a strategy until it looks brilliant on the past, which teaches you about the past. The honest test is paper mode running forward on data that had not happened when you wrote it. That is why every bot here starts in paper and stays there until you deliberately arm it.
Open Bots in the left nav.
Bots run on your machine
This is the first thing to understand, and it cuts both ways.
Your strategies execute in Pairlens, on this computer, not on a server we operate. That is what keeps your keys and your logic yours. The tradeoff is real: quit Pairlens or let the machine sleep and your bots stop with it, and an open position is left unwatched until you come back.
The Bots page has a Keep this computer awake toggle that blocks idle sleep while a bot is on. It is desktop-only, because a browser tab cannot make that promise. Turn it on before you leave a bot running.
If your strategy declares a stop_loss, and your venue supports exchange-native
trigger orders, that protection can rest at the exchange and survive the app
closing. Everything else needs Pairlens running.
Creating a bot
Four steps.
1. Strategy
Pick the script the bot will run. The step offers three ways in, so it never dead-ends:
- Your strategies. Scripts declaring
meta = strategy(...), because only those say when to enter, when to exit, and how much to commit. Anindicator(...)script draws and nothing more, so there is nothing for a bot to execute. Scripts you have not run yet are listed but not selectable: run them once in the workbench first. - Ready-made strategies. The shipped strategies (EMA cross, RSI reversion, breakout) can be created right here. Picking one writes the script into your workbench and selects it, code and all, yours to edit afterwards.
- Write your own. The link at the bottom opens Indicators & Strategies, the Python workbench where strategies are written. A strategy script there has a Deploy as bot button that lands back in this dialog with the script preselected.
There is also a fourth way that skips the dialog entirely: the assistant, one tap away on the sparkle button above the bot list or on its dock from any page. “Deploy my breakout strategy on OKX BTC-USDT, 1h” creates the bot in one exchange.
It can start further back than that. A bot is a strategy on a market, so asking for one you have no script for (“a bot that buys pullbacks in an uptrend on the 4h”) gets both: it writes the strategy, validates it, backtests it, and deploys the result. When the code needs real work it opens the workbench and keeps going there, in the same conversation.
Anything it creates arrives in paper mode and switched off, exactly as if you had walked the steps yourself, and it can never arm a bot. It can also rename bots, tune strategy params, and set guards on your behalf when you ask, and it asks you back when a choice is yours to make. See Build with AI for how the assistant works and which AI providers it uses.
2. Market
Choose a venue, a pair, and a timeframe. The bot evaluates on that pair’s closed candles at that timeframe. A 1h bot makes at most one decision an hour.
3. Sizing
How much the bot commits when the strategy signals an entry:
| Mode | Meaning |
|---|---|
| Percent of equity | A share of the account, per entry |
| Fixed quote amount | A constant amount in the quote asset |
| Fixed base amount | A constant amount in the base asset |
4. Guards
Limits enforced outside the strategy, by the bot runtime. Leave any blank for no limit.
| Guard | What it does |
|---|---|
| Daily loss cap (%) | Stops the bot once today’s realized losses reach this share of equity |
| Max trades per day | Stops the bot after this many entries in a rolling 24 hours |
| Max position (quote) | Never hold more than this notional |
| Cooldown bars | Wait this many closed bars after a losing exit before re-entering |
| Max losing streak | Stops the bot after this many losing trades back to back |
Cooldown bars and max losing streak are the two that most often turn a theoretically profitable strategy into a survivable one. A strategy that is wrong is usually wrong several times in a row.
These are on top of your account-wide risk guardrails, not instead of them.
Paper first
A new bot starts in Paper mode. Fills are simulated locally against the same closed candles the strategy sees, using the fee and slippage the script declared. Nothing reaches the venue.
Leave it running for a week. The trade ledger and event log fill in, the P&L chart traces the running total as each round trip closes, and you get to judge the strategy on data that had not happened when you wrote it.
Going live
Flip Live trading in the bot’s settings and an arming dialog appears. It tells you which venue and which credential will be used, states plainly that the bot will place real orders from now on with no confirmation per trade, and repeats the machine caveat.
Then you type ARM LIVE to confirm.
If no API credential is stored for that venue, the dialog says so and sends you to Accounts rather than letting you arm something that cannot trade.
Turning live trading back off stops the bot.
Watching a bot
The bot detail view has three tabs plus a header.
Header. Status (Running, Warming up, Stopped, Error, or Halted by a guard), current position with unrealized and realized P&L, last bar, last price, venue, strategy, and timeframe.
Trades. The ledger: side, size, entry, exit, exit reason, mode, and P&L. Above it, a summary strip with closed trade count, win rate, value traded, average per trade, best, and worst.
Events. Signals, orders, and guard blocks, timestamped. When a bot halts, this is where it says which guard stopped it.
Settings. Strategy inputs, sizing, guards, the live toggle, and rename or delete.
Three charts sit alongside: realized P&L over time, value traded, and trades per day.
Halted, and re-arming
When a guard trips, the bot halts rather than quietly continuing. Its status reads Halted by a guard and the event log names the guard. Fix the cause, or accept it, then Re-arm to resume.
That is a deliberate speed bump. A bot that trips its losing-streak guard and restarts itself is not a guard, it is a delay.
Managing several
The sidebar lists every bot with its mode and status, and the header summarises how many are deployed and how many are on. Stop all kills every running bot at once, which is the button you want when something is happening in the market that your strategies were not written for.
Duplicating a bot copies its configuration, which makes running the same strategy across three pairs a thirty-second job.
The bot you have open is in the address, as /bots?bot=<id>. Copy it and you
have a link straight to that deployment, the back button walks between the bots
you looked at, and the assistant knows which one you mean when you ask about
“this bot”.
Deleting
Deleting a bot removes it along with its trade ledger and event log. Any position it holds at the venue is left untouched, so close the position first if you meant to close the position.