AI trading bot
An AI trading bot you can read and test
An AI trading bot is software that uses AI to decide or design trades, and the kind worth trusting shows you its rules before any money moves. Keel is an AI trading bot builder: you describe a strategy, the AI builds it from blocks you can read, and you backtest it with fees, slippage and funding and paper trade it before it runs from your own account on Hyperliquid. Starting is free.
By Keel Research Team
Updated October 18, 2026A bot from the Keel library
Aug 2024 to Sep 2026Sharpe
1.15
Total return
+213.4%
Worst drawdown
−50.2%
Trades
1,141
Its default settings, after trading fees, slippage and funding.
What an AI trading bot is
Two very different things share the name
Most products sold as an AI trading bot are a black box. A model, or a set of rules the seller calls AI, watches the market and places trades on your account. You see the trades after they happen. You rarely see why, and you can't change how it decides.
The other kind uses AI earlier. You tell the AI what you want the bot to do, and it writes the rules: which coins, what signal, when to get out, how much to put in each position. Those rules then run exactly as written, the same way every time. You can read them, change them and test them on past data before they touch your money.
Both are automated, and both can lose money. The difference is what you can check. With a black box you are trusting the seller. With rules you can read, you are trusting a backtest you can rerun yourself.
| Black-box AI bot | AI-built rule (Keel) | |
|---|---|---|
| What the AI does | Decides trades while it runs | Writes the strategy before it runs |
| What you can read | The trades, after the fact | Every rule and setting, before it trades |
| Can you test it first | Only if the seller publishes a test | Yes, on real price and funding history, with costs |
| When it has a bad month | Hard to tell a bad market from a bad model | You can see which rule did what and change it |
How it works on Keel
From a sentence to a bot that runs
You don't need to code, and you can stop at any step. Most people spend their time testing versions until one holds up.
- 01
Describe the bot
Write what it should do the way you would tell another trader: hold the strongest coins, buy dips in an uptrend, get out when the trend breaks. You can also start from a strategy in Keel's library and ask for one change.
- 02
See it as blocks
The AI builds the bot from Keel's 251 building blocks: market data, indicators, filters, exits and position sizing. Each block shows its settings on a canvas. The AI can only use blocks that exist, so it can't invent an indicator.
- 03
Backtest it
Keel runs the bot over real price and funding history and charges trading fees, slippage and funding as it goes. You get the equity curve, the worst drawdown and every trade, and you can ask the AI what went wrong in a bad month.
- 04
Paper trade it
Run the same bot on live prices with simulated money. You need no wallet and no exchange account. Watching it on new data is the best check that the backtest wasn't luck.
- 05
Run it live
Connect the wallet you use on Hyperliquid and sign once. The bot trades from your own account through an agent wallet that can place trades but can never withdraw, and you can pause it or revoke access at any time.
A bot you can read
What a bot built from readable rules looks like
This is a trend bot from Keel's public library, the sort of bot people ask an AI to build. It holds liquid coins while they trend and steps aside when they don't. Below it is shown exactly the way Keel shows every strategy, with a code view one click away.
Start at the top. Daily prices come in for the most traded perps. Three checks must all pass before a coin is bought: a faster moving average sits above a slower one, recent momentum is positive, and the funding a long pays is below a cap. A trade manager then holds the position until the averages cross back or price drops far enough from its high to hit a trailing stop sized to the coin's volatility. The last blocks split the capital equally across whatever qualifies, with a cap on any one coin.
Configuration
Globalsconfig
target_timeframe1dbar_offset12h
UniverseassetsTop by Volume · HL Perps · 20 assets20 resolved
Executionorders
rebalanceOn Change
Factories
ema_crossfactory
Pipeline7 steps
PriceDataLoaderData Loader
ohlcv_1dstore
Parallel2 branches
entries
Parallel2 branches
const
ConstantForecastForecast Mapper
value1
gates
Parallel3 branches
trend_up
ema_crosscall
AboveThresholdFilterSignal Transform
threshold0
momentum_agrees
ROCIndicator
period14
AboveThresholdFilterSignal Transform
threshold0
funding_ok
FundingDataLoaderData Loader
use_cachetrue
BelowThresholdFilterSignal Transform
threshold0.0000625inclusivetrue
MaskAndSignal Composer
ApplyMaskSignal Composer
score_signalconstfilter_signalgates
bot_entrystore
exit_flip
ema_crosscall
NegateTransformSignal Transform
ThresholdCrossSignal Transform
upper0modelong_only
bot_exit_flipstore
TradeManagerPosition Manager
entriesbot_entrypricesohlcv_1d
Parallel2 branches
trend_flip
bot_exit_flipload
AboveThresholdFilterSignal Transform
threshold1inclusivetrue
ExitPosition Manager
atr_trail
TrailingStopPosition Manager
atr2.5period14
ExposurePosition Manager
EqualWeightSizerPosition Sizer
target_leverage1max_weight0.2
Globals(target_timeframe="1d", bar_offset="12h")
Universe(mode="top_volume", top_n=20, market="perp", resolved=["BNB", "BTC", "DOGE", "ETH", "FARTCOIN", "GRAM", "HYPE", "LINK", "LIT", "NEAR", "PUMP", "SOL", "SUI", "VVV", "WLD", "XPL", "XRP", "ZEC", "kBONK", "kPEPE"], resolved_at="2026-07-06T20:20:32.738649+00:00")
Execution(rebalance="on_change")
ema_cross = Pipeline([
{
"fast": [EWMA(window=10)],
"slow": [EWMA(window=30)],
},
Crossover(fast_key="fast", slow_key="slow"),
], name="ema_cross")
Pipeline([
PriceDataLoader(),
Store("ohlcv_1d"),
{
"entries": [
Pipeline([
{
"const": [ConstantForecast(value=1.0)],
"gates": [
Pipeline([
{
"trend_up": [ema_cross, AboveThresholdFilter(threshold=0.0)],
"momentum_agrees": [ROC(period=14), AboveThresholdFilter(threshold=0.0)],
"funding_ok": [FundingDataLoader(use_cache=True), BelowThresholdFilter(threshold=0.0000625, inclusive=True)],
},
MaskAnd(),
], name="gates"),
],
},
ApplyMask(score_signal="const", filter_signal="gates"),
Store("bot_entry"),
], name="entries"),
],
"exit_flip": [ema_cross, NegateTransform(), ThresholdCross(upper=0.0, mode="long_only"), Store("bot_exit_flip")],
},
TradeManager(entries="bot_entry", prices="ohlcv_1d"),
{
"trend_flip": [Load("bot_exit_flip"), AboveThresholdFilter(threshold=1.0, inclusive=True), Exit()],
"atr_trail": [TrailingStop(atr=2.5, period=14)],
},
Exposure(),
EqualWeightSizer(target_leverage=1.0, max_weight=0.2),
], name="ai_trend_bot_crypto")
Nothing in this bot calls an AI while it runs. On Keel the AI's job is writing a bot like this, and yours is checking it. The result at the top of the page is this bot on its default settings, after trading fees, slippage and funding. Stops and exits are checked at the close of each bar, so a fill can land past the stop level in a fast move.
See every setting tested and the monthly returnsFork this bot into your account
Lands a copy in your account to change and backtest. Nothing trades until you start it.
AI trading signals
AI trading signals, and what Keel does instead
An AI signal service sends you alerts: buy this coin, sell that one, sometimes with an entry and a stop. You still decide whether to act and place the trade yourself, after the alert arrives. Ask any service for a full record of its past signals after fees before you pay.
Keel doesn't sell signals. What it gives you is the rule behind a signal: the conditions that would fire it, written as blocks, backtested over past data and then run on your account without you watching for alerts. If you already follow a signal you like, describe its logic to the AI and test whether it held up.
A good test for any signal, from Keel or anyone else: could you write down the rule that produced it? If yes, you can backtest it. If no one can, you are following a guess.
Free AI trading bot
What you can do on the free plan
Starting is free. The free plan lets you build and backtest strategies, paper trade one strategy and run one live, with a 5 basis point fee on live trades. Paid plans start at $29 a month and lower that fee. The pricing page has the full limits.
That is enough to build a bot, test it properly and watch it on paper before you spend anything. Paid plans make sense once you are testing many versions a week or running more than one strategy at a time.
Before you trust one
Five things to check on any AI trading bot
These apply to every bot, ours included. Each has a plain answer you can find before you connect money.
How often it trades
A bot that trades many times a day pays fees and slippage on every trade. Plenty of bots that look clever lose most of their edge to costs. Count the trades in the backtest and check that costs were charged on each one.
Whether costs were in the test
Fees, slippage and, on perpetual futures, funding. Leave any of them out and the result looks better than it will trade. Ask which costs the backtest charges by default.
Whether you can rerun the backtest
A screenshot of returns proves little. You want to rerun the test yourself, change a setting and see how much the result moves. A bot whose numbers fall apart when one setting changes was probably fitted to the past.
Who holds the money
Some bots ask you to deposit with them. Others trade through a key or a signed permission on an account you own. Know which, and whether the bot can withdraw.
What the AI actually does
Picking trades in the moment, writing rules or just explaining results are very different jobs. A bot should say plainly which one its AI does.
FAQ
Questions about AI trading bots
Is there a free AI trading bot?
Yes. On Keel you can build an AI trading bot, backtest it and paper trade it on the free plan. Starting is free. The free plan lets you build and backtest strategies, paper trade one strategy and run one live, with a 5 basis point fee on live trades. Paid plans start at $29 a month and lower that fee. The pricing page has the full limits.
Do AI trading bots work?
Some make money and many don't, and the AI label tells you little either way. What matters is the rule the bot follows and whether it held up after costs. Our study of whether crypto trading bots work looks at that across the library.
Can AI predict the market?
Not reliably. When language models were each given real money to trade crypto perps in a public contest, most of them lost, and much of the damage came from trading too often. AI is more useful for writing and testing a rule than for guessing the next move.
Are AI trading signals worth paying for?
Only if the seller shows a full record of past signals after fees, and most don't. A rule you can backtest yourself tells you more than a feed of alerts. Keel doesn't sell signals; it runs rules you have tested.
Do I need to code?
No. You describe the bot in plain words and the AI builds it from blocks you can read and change on a canvas. There is a code view if you want it, and you can also build from Claude or ChatGPT.
Which markets can it trade?
Keel runs on Hyperliquid's crypto perpetual futures today, such as BTC, ETH, SOL and HYPE. Stock and index perps are HIP-3 markets, and Keel doesn't support them yet. Support is coming soon.
Who holds my funds?
You do. Your money stays in your own Hyperliquid account. Keel trades through an agent wallet you approve, which can place trades but can never withdraw or transfer funds, and you can revoke it from Keel or on Hyperliquid at any time.
Can I test it without money?
Yes. Every bot can be backtested on past data and paper traded on live prices with simulated money. Paper trading needs no wallet or exchange account. How Keel works explains how the backtest, paper and live runs use the same strategy.
Build a bot you can read
Describe it, see the rules, test it with real costs, and decide whether it trades.