When people hear “trading bot,” they picture a little script that watches a price and fires an order. That's the part of TradeKarma most people would notice, and it's honestly the least interesting part. The bot is just the mouth. Everything that makes it worth trusting with real money sits behind it.
It starts with data, because nothing downstream is better than what's coming in. TradeKarma pulls live futures data straight from Rithmic — every tick, in real time — and it doesn't just glance at it and forget. It writes all of it down. We keep our own history in a time-series database, more than a year of it now, so we're never at the mercy of some vendor's ten-day lookback or a feed that quietly changes. That archive is the thing you can't buy back later. Every backtest, every “would this have worked,” leans on data we captured ourselves and can trust.
From that raw stream, the system builds understanding. It rolls ticks into candles, then into higher timeframes. It computes the stuff that actually moves markets but doesn't show up on a basic chart — cumulative delta, volume profile, where price is being accepted or rejected. Once a day it works out the levels that matter: yesterday's high and low, where Asia traded, where London traded, and caches them so every strategy is looking at the same map.
Only then does the “bot” part happen, and even that isn't one thing. It's an engine that runs several strategies at once, each just a config file, each watching for its own setup. When one fires, the signal fans out to multiple prop accounts in parallel, brackets and stops and all. And here's a detail I'm quietly proud of: the thing that tracks the trade in our head and the thing that talks to the broker are kept separate on purpose. The tracker can believe a trade is open a half-second before the broker confirms it — and if the broker never does, a watcher cancels the ghost so we never manage a position that doesn't exist.
Wrapped around all of that is the part nobody sees until something breaks. There's a supervisor that babysits the whole process, watches for a heartbeat, and restarts it if it dies. There's a dashboard that shows live P&L, open trades, account health, and a journal of everything that's happened. There are machine-learning layers riding along — scoring signal quality, watching for exits, reading market regime — all in an advisory seat for now, quietly building the track record they'd need before we ever let them touch a trade. There's even a layer pulling options data to sense where the big dealers are positioned, just accumulating history until it's ready to earn its keep.
The whole thing points at one goal: it should start itself, feed itself, trade, manage, recover from a dropped connection, and keep going without a human hovering over it. Every piece gets judged by whether it moves us closer to that or adds a chore.
So no, it's not a bot. It's a small operation that happens to run on one machine — data desk, quant, execution, risk, and night-shift babysitter, all wired together.
Related: How We Actually Build a Strategy →