I built an autonomous trading bot, handed it a $100 crypto wallet, and told it to trade memecoins on Robinhood Chain by itself: find tokens, buy them, manage the trades, sell them. I controlled the whole thing from Telegram, and I built and ran it entirely from my phone on a cheap always-on cloud box.
The $100 and the ~$5 per trade were deliberate: small on purpose, real money but cheap enough to be a test bed. The whole point was to tweak the strategy against live markets without betting anything I'd miss, tuition, not a bankroll.
Let me spoil the ending: it lost money. Down about a third. But it was one of the most educational things I've shipped, and the failure is more interesting than a win would've been. Here's the honest journey, the full setup so you can try it too, and where I'm taking it next.
The setup: a bot you drive from Telegram
The whole system runs on a €5/month Hetzner devbox, always on, so the bot trades 24/7 whether my laptop is open or not. I built the entire thing from my phone with Claude Code over SSH, which meant I could tweak the strategy from a café, ship it, and watch the results come into Telegram minutes later. That feedback loop is the whole point.
The architecture is simple:
- A scanner polls the chain every ~45s (new launches, trending, top-volume pools), scores each token, and decides what's worth buying.
- An on-chain executor places the actual swaps through the chain's Uniswap forks, paying in native ETH and unwrapping proceeds back.
- A risk manager rides each position with a hard stop-loss, a laddered take-profit, and a trailing stop.
- Telegram is the cockpit: every buy, sell and take-profit pings my phone, and I can check stats or sell a position with a tap.
Every action shows up like this:
DATABEAR: bought a deep-liquidity climber, banked a third at +30%, trailed the rest out for a clean +35% on the trade.
What made it hard: the scams
Here's the thing nobody tells you about trading tiny memecoins: most of the chain is a minefield of scams, and they get more creative than you'd believe. Two-thirds of my losses had nothing to do with my strategy being wrong, they were tokens designed to steal from you.
Honeypots
You can buy, but you can't sell, the token's code blocks it. Your money's just stuck. I learned this the hard way early, got trapped in one and watched the money sit there, unsellable. So preventing scams was baked into the plan from the start: right after that first hit I built a pre-buy simulation that fakes owning the token and simulates a sale on-chain before spending a cent. If the sale reverts, we don't buy. From then on it blocked the classic ones cleanly.
Fee-on-transfer tokens
These tax every transfer, which breaks the exact-swap math and makes the buy itself revert. The simulation catches these too, we just skip them and move on.
Sweepers (the nasty one)
A token called SCATMAN passed both my buy and sell simulations, because in a fresh simulation it genuinely is sellable. But look at the on-chain trail: I bought 10,428 tokens at 06:40:45, and three seconds later a malicious hook forwarded every one of them out of my wallet to the scammer's address. The theft only fires when real tokens land in a real wallet, so no pre-buy check can see it. That's an unblockable-by-design attack, and it cost the price of the position.
I added a delayed balance re-check that catches sweepers a minute after buying, it can't recover the money, but it flags them instantly and never touches that token again.
The slow chain
Robinhood Chain has ~101-second blocks. That means every order lands ~2 minutes after you decide, so your fill is at a different price than you saw, and a trailing stop reacts to a dump two minutes late. On volatile memecoins that lag is brutal, you buy a pop and it's already reversed by the time you're filled.
What worked, and what didn't
The engineering worked beautifully. Honeypot defense, multi-fork routing to reach the deepest pool, on-chain pricing instead of a laggy API, laddered take-profit, an adaptive trailing stop that tightens as a winner runs, a full live dashboard. All of it does exactly what it should. When the bot caught a real runner, it milked it perfectly.
What didn't work is the part no amount of engineering fixes: the edge. On this chain, the tokens that pass every quality filter still mostly pop ~12% and reverse. Whether I chased momentum or waited for pullbacks, entries kept landing near -20%. The market itself is hostile, thin, scam-ridden, and slow. You can build a perfect machine and still be fishing in a bad pond.
The dashboard
Everything the bot did was published to a self-updating dashboard: total P&L split into trading vs scams, an equity curve, every trade with a win/loss badge, and a plain-English explainer of exactly how it entered and exited. Here's the final state, the honest scoreboard, split so you can see the strategy apart from the scam tax:
Want to build your own?
The blueprint is genuinely simple, and it's a fantastic project because you get a tight feedback loop with real stakes. If you want to try:
- Get an always-on box so it trades 24/7, here's the €5/month devbox setup. Bonus: you can build and babysit the whole thing from your phone.
- Make Telegram your cockpit. Create a bot with
@BotFather, and have your script send a message on every trade and answer commands like/statsand/sell. Being able to see and steer from your pocket is what makes it fun. - Defend before you trade. Simulate a sell (and a buy) on-chain before spending. Assume every token is a scam until proven otherwise, that lesson cost me real money.
- Split your scoreboard. Track scam losses separately from trading losses. It's the difference between "my strategy is broken" and "this market is a swamp."
- Start with $100. Enough to be real, cheap enough to be a tuition bill.
So… now what?
Not a success. But I'd do it again in a heartbeat, and I'm not stopping. The takeaway is clear: the machine is good, the pond is bad. So I'm changing ponds.
The next bot is deliberately the opposite of this one: one real, deep-liquidity asset (HYPE on Hyperliquid), one dead-simple strategy: a bounded grid that buys dips and sells small bounces, with hard caps so a downtrend can't blow it up. No scams, no thin pools, no 2-minute blocks. It's already running in paper mode, proving itself against real prices before a cent goes in.
Follow along. I'll write up the HYPE bot and its results next. If you build your own, tell me how it goes.