Trading bots are here

The final rung of the Research → Trade funnel is here: trading bots. A bot runs one of your saved strategies hands-free against today's live session — the same entry and exit logic your backtest measured, executed in real time whether you're watching or not. It's the bridge between "the backtest looks good" and "it trades while I watch". And like everything else on the platform, it's simulated money against the live market — nothing here touches a real brokerage.

A bot is a dedicated trading account with three things pinned to it. First, the exact strategy that produced the backtest you evaluated — form-built or written with the AI assistant — so what runs live is precisely what you tested. Second, a risk policy: per-bot runaway tripwires like caps on open positions and orders, a stale-data threshold, and an error budget that pauses a strategy that keeps erroring. Third, a continuous balance that carries from one session to the next and compounds, like any trading account.

Running one is a single tap to start today's session — or flip on automatic start and the platform opens it for you at the market open, at most once per trading day, only on days the strategy actually enters. Through the day you can watch everything live: orders, positions, per-second P&L, and a decision log showing what the strategy saw and chose, tick by tick. And you can always step in and stop the session. Stopping a session stops the strategy, not the trade: whatever the bot holds stays open, and you choose the exit — liquidate the positions smartly (grouped limit orders worked patiently toward a fill, so you keep the spread), liquidate them aggressively (a market order per position, immediate), or hold them to the close and let the day settle exactly as it would have.

Bots also trade smarter than a naive robot would. Backtests fill a mid-priced limit order instantly; the live market is not so generous — an order resting at the mid tends to sit unfilled while the market drifts. So every bot order goes in at a spread-aware starting price and, if it hasn't filled, is re-priced one tick at a time toward a fill — entries patiently, exits with urgency, and a stop-loss faster than a take-profit, because a losing position is more urgent to leave than a winning one. All of it is on by default and tunable per bot.

Automation amplifies bugs, so every bot session is watched for exactly that: if a session starts firing orders in a runaway loop, the platform halts it immediately, cancels its working orders, and settles the day cleanly. Those limits are set well above anything a healthy strategy does — it's purely a brake on a bot that has clearly gone wrong.

Update, July 31: the fill model described above has changed. Backtests no longer fill a mid-priced limit order instantly — a fill takes three consecutive seconds everywhere, and a resting limit fills at its own price. See More realistic fills. Everything else in this post still stands.

Everything above works from the app and over the Bots API alike, and the new Learn guide covers it all in depth. One piece of advice from that guide worth repeating here: earn the bot. Backtest the strategy over the full history, trade its best and worst days by hand in Practice, run it live yourself for a while — then hand it over. Automation amplifies whatever you give it, including flaws.