What's the actual difference between on-chain and CEX bots?
A CEX bot talks to a centralized exchange (Binance, Bybit, Coinbase) over a REST and WebSocket API. Your funds sit in the exchange's wallet; the bot just sends order instructions authenticated by an API key. The matching engine runs on the exchange's private servers.
An on-chain bot interacts with smart contracts on a blockchain. It signs transactions with your private key and submits them to a node, either to an automated market maker like Uniswap or a perp DEX like Hyperliquid, GMX, or dYdX. Nobody custodies your money but you. If you want the mechanics of how these connections work, our piece on trading bot APIs covers both models.
The split matters because four things change with it: who holds your money, what you can verify, how fast you execute, and what a round trip costs.
Custody: who holds your money while the bot trades?
This is the biggest fork in the road. On a CEX, the exchange holds your assets. You get speed and simplicity, but you inherit counterparty risk. FTX, Celsius, and a dozen smaller failures all held customer funds that vanished. An API key with withdrawal permission is also a fat target, which is why we push readers hard on securing bot API keys and disabling withdrawal scope entirely.
On-chain, you keep custody. The bot signs orders with a key you control, and funds either stay in your wallet or sit in a transparent, audited contract you can exit at any block. There's no exchange that can freeze your account or halt withdrawals.
But self-custody moves the risk, it doesn't delete it. Now you carry:
- Smart-contract risk — a bug or exploit in the protocol can drain the pool.
- Key management risk — lose the private key or leak it and the funds are gone, no support desk to call.
- Approval risk — a malicious token approval can let a contract sweep your wallet.
Rule of thumb: if the bot needs withdrawal rights on a CEX, or an unlimited token approval on-chain, treat that as the real attack surface, not the trading itself.
How do fees compare, and where do the hidden costs hide?
Headline fees favor centralized venues and modern perp DEXs; legacy AMM swaps are the expensive outlier once you count everything.
On a CEX, standard spot fees run around 0.10% (10 bps) maker and taker, dropping toward 1-2 bps at high VIP volume tiers. Binance perpetual futures sit near 2 bps maker and 5 bps taker at the base tier. Hyperliquid, a perp DEX, publishes some of the lowest fees anywhere at roughly 1.5 bps maker and 4.5 bps taker, with staking and volume discounts on top. So "on-chain" is not automatically pricier.
The expensive on-chain case is AMM spot swaps. A Uniswap trade pays the pool fee (commonly 0.30%, or 30 bps), plus network gas, plus price impact and slippage on anything but the deepest pools. Gas is a fixed dollar cost per transaction, so it punishes small orders brutally: a $20 gas fee on a $500 trade is 400 bps before you've traded anything. That kills most high-frequency and arbitrage strategies on Ethereum L1.
| Cost dimension | CEX bot | On-chain bot |
|---|---|---|
| Trading fee (perps) | ~2 / 5 bps maker/taker | ~1.5 / 4.5 bps (perp DEX) |
| AMM swap fee (spot) | n/a | ~30 bps + slippage |
| Per-transaction gas | None | Fixed $ cost, hurts small orders |
| Custody / withdrawal | Exchange holds funds | You hold keys |
| Failed-order cost | Free (rejected) | Gas still burned on revert |
Two hidden on-chain costs bite bot traders specifically. First, reverted transactions still cost gas, so a bot that spams orders in volatile conditions bleeds fees on trades that never fill. Second, on public mempools your pending transaction is visible, inviting MEV bots to front-run or sandwich it. For a full cost picture across both worlds, see our breakdown of what trading bots actually cost.
Latency and execution: does on-chain lag actually matter?
For fast strategies, yes, a lot. A CEX matching engine confirms in single-digit milliseconds, and colocated setups push that lower. WebSocket order-book feeds let a bot react to depth changes in near real time. If your edge depends on speed, a centralized venue is hard to beat.
On-chain, you are bounded by block time and finality. General-purpose L1s like Ethereum settle roughly every 12 seconds, which is an eternity for a market-making or scalping bot. Purpose-built perp DEXs close the gap dramatically: Hyperliquid runs its own high-throughput chain with sub-second blocks, which is why serious bot activity has migrated there rather than to AMMs. Our guide to Hyperliquid trading bots goes deeper on that ecosystem.
Latency also shapes which strategies are even viable:
- Market making and arbitrage — need low latency; strongly favor CEX or a fast perp DEX.
- Grid and DCA bots — tolerant of seconds-level delay; run fine on-chain.
- Copy trading — depends on how fast you mirror; on-chain copy trading on perp DEXs is popular precisely because positions are public, covered in copy trading on perp DEXs.
Transparency: what can you actually verify?
On-chain is radically more transparent, and for some traders that alone decides it. Every order, fill, liquidation, and funding payment is written to a public ledger. You can independently audit a protocol's total open interest, verify a vault's real track record instead of a screenshot, and confirm your own positions without trusting a dashboard.
A CEX is a black box by comparison. You see your account and a public trade tape, but the internal order flow, matching logic, and the exchange's own solvency are opaque. Historically that opacity has hidden wash trading, inflated volumes, and, in the worst cases, insolvency that customers only learned about at collapse.
This transparency is why on-chain trader-selection metrics can be trusted more than CEX leaderboards: the numbers are verifiable rather than marketed. The flip side is privacy. Every move your bot makes is public forever, so anyone can reverse-engineer your strategy from the chain.
Which should you pick?
Match the venue to the strategy, not the other way around. There's no universally better choice, and running both is common.
Lean CEX when:
- Your strategy is latency-sensitive (market making, HFT arbitrage, tight scalping).
- You trade small size where fixed gas would dominate returns.
- You want the simplest possible setup and the deepest liquidity.
Lean on-chain when:
- Self-custody is non-negotiable and you refuse counterparty risk.
- You want verifiable, auditable execution and vault performance.
- You're trading perps on a fast DEX where fees now rival or beat centralized venues.
Whichever side you land on, the uncomfortable base rate doesn't change: most retail bots lose money after fees, funding, and slippage regardless of venue. Custody and transparency reduce specific risks; they do not create edge. Size positions conservatively and stress-test any unattended bot before you trust it with real capital. This site, trading.bot, exists to compare these mechanics neutrally, not to sell you either model.
Frequently asked questions
Are on-chain trading bots safer than CEX bots?
Neither is strictly safer; the risks differ. On-chain bots remove exchange counterparty and withdrawal-freeze risk because you keep custody, but add smart-contract, key-management, and MEV risk. CEX bots simplify security but expose you to exchange insolvency and hacked API keys. Match the risk profile to how much you trust the venue.
Do on-chain bots have higher fees than CEX bots?
Not always. AMM spot swaps are expensive, around 30 bps plus gas and slippage, which punishes small or frequent trades. But modern perp DEXs like Hyperliquid charge roughly 1.5 bps maker and 4.5 bps taker, competitive with or cheaper than most centralized futures venues. The venue type matters more than the on-chain vs CEX label.
Why is latency worse on-chain?
On-chain execution is bounded by block time and finality. General L1s like Ethereum settle roughly every 12 seconds, far too slow for scalping or market making. Purpose-built perp DEXs run dedicated high-throughput chains with sub-second blocks, which narrows the gap enough that many bot strategies now run on-chain successfully.
Can I run the same bot on both a DEX and a CEX?
Yes, and many traders do. A CEX connection uses REST and WebSocket APIs with an API key, while an on-chain connection signs transactions with a wallet key and submits them to a node. The trading logic can be shared, but order routing, fee handling, and latency assumptions must be tuned separately for each venue.
Is on-chain trading more transparent than a centralized exchange?
Yes. Every order, fill, liquidation, and funding payment on-chain is recorded on a public ledger you can audit independently, including a vault's real track record. A centralized exchange only shows your account and a public trade tape; its internal matching and solvency stay opaque, which has historically hidden wash trading and insolvency.