KimchiBot

A walkthrough of how I built a Claude AI trading bot end to end — strategy logic, ccxt integration with Binance Futures, stop loss handling, coin rotation, and the bugs I had to fix along the way.

I wrote the entry/exit logic of my Claude AI trading bot in about an hour. The strategy itself wasn’t the hard part — the structure around it was. Here’s what I had to build after to make a Claude trading bot survive live data.

My bot’s first three weeks of live trading were rough. A deep early drawdown forced me to think hard about whether the strategy or the structure was at fault. Here’s what I changed and why.

A breakdown of how the bot performed in April 2026. Trade counts, drawdown, fee load, and what the per-coin and per-day data revealed about the strategy.

I built 6 trading bots. 4 of them are dead. Grid bot, RSI scalper, momentum bot, lead-lag bot — each died for a different reason. Here’s what killed them.

My bot tried to close a SHORT position. Binance rejected the order. The bot thought it succeeded. The position bled -121U before I woke up.

Some coins hit your stop loss on every single FVG trade. I built a filter that bans them automatically. The results were dramatic.

My FVG bot has a win rate around 33%. That sounds bad until you do the expectancy math. Here’s why win rate is one of the most misleading metrics for evaluating a strategy.

I tested my Fair Value Gap bot on 4 quarters of data it had never seen. 3 out of 4 were profitable. Here are the exact numbers.

My bot tried to close a position. The exchange said no. Three times. Here’s what happens when market close fails and how to survive it.

My bot doesn’t trade fixed coins. Every 3 hours it scans all Binance Futures pairs, filters by volatility, backtests each one, and picks the top 8. Here’s the full algorithm.

If your live bot doesn’t match your backtest, your backtest is fiction. Here’s the comparison tool I built and the bugs it caught.

My bot left a stop loss order on the exchange after closing a position. Hours later, it triggered and opened an unintended trade. Here’s what caused it and how I fixed it.

I gave Claude Code my trading data, my prior failures, and my constraints, and asked it to suggest a strategy from scratch. Here’s how the experiment was set up and what the results actually told me.

STOP_MARKET is the default stop loss on Binance Futures. It’s also silently eating your profits through slippage. Here’s exactly how much it costs and how to switch to STOP_LIMIT.

A complete beginner’s guide to building a crypto trading bot with Python and ccxt. From zero to a working bot on Binance Futures.

Everything you need to know about using ccxt with Binance Futures in Python. Connection, orders, positions, stop losses, and all the gotchas.

A developer from Korea built 6 crypto trading bots using Claude Code. 4 failed. Here’s the honest story of building a Claude AI trading bot — and a guide so you can build one too.

Overfitting is the #1 killer of trading bots. Here’s a practical checklist to detect and prevent it, based on real experience.

Grid bots look perfect on paper. Buy low, sell high, automatically. Here’s why I killed mine after a week.

Most trading bot tutorials show you the strategy but not the structure. Here’s how to organize a production-ready trading bot in Python.

My momentum bot showed incredible returns in backtests. Then I learned about overfitting — the hard way.

I’ve built 6 trading bots and made every mistake possible. Here are the 10 most expensive ones so you don’t have to repeat them.

My bot showed +500U in backtests. Live trading showed +200U. Here’s every gap I found and how I closed them.

Claude AI trading bot vs ChatGPT trading bot: an honest comparison after building both. What each AI is good at, what each fails at, and which one I actually use for live trading.

How I built a trend following crypto bot in Python: signal logic, exits, position sizing, and the design choices that came out of killing four earlier strategies.

I went from market order SL to exchange-side STOP_LIMIT and it changed everything. Here’s every mistake along the way.

My bot doesn’t trade the same coins forever. Every 3 hours, it picks the best ones automatically. Here’s the algorithm.

I spent weeks building a WebSocket price feed. Then I realized I didn’t need it at all.

After trend-following, I built a completely different bot based on Fair Value Gaps. It passed a full year of out-of-sample testing.

My bot was missing every trade for the first 9 hours of each day. The cause? Mixing UTC and KST in one line of code.

A practical guide to building a Claude AI trading bot. What Claude Code is great at, what it’s terrible at, and how to get the best results when building a crypto trading bot with AI.

Forget win rate. A 35% win rate can make you rich. A 70% win rate can bankrupt you. Here’s the math.

A collection of Binance Futures API quirks that aren’t in the documentation. Each one cost me hours.

One bot follows trends. The other trades mean reversion. Together, they cover each other’s weaknesses.

My bot was running twice. Two instances, same account, doubling every order. Here’s how PID lockfiles saved me.

When a trade moves +4% in the first 5 minutes, you need special handling. Here’s my surge detection system.

Your bot will crash. At night. On a weekend. With open positions. Here’s how to make sure it recovers gracefully.

Grid search found parameters that turned -82U into +343U. Here’s how I made sure it wasn’t just overfitting.

Between backtest and live trading, there’s a step most people skip. It’s the step that catches the bugs that matter most.

Binance offers up to 125x leverage. I use 3x. Here’s the math behind that boring decision.