← All articles

How trade copiers actually work (and where the milliseconds go)

Every trade copier does the same three things: detect that your leader account filled, decide what each follower should do, and submit those orders. The differences that matter — speed, reliability, which accounts it can reach at all — come from where those three steps run.

The three architectures

Platform add-ons (the NinjaTrader 8 ecosystem — Replikanto is the best-known) run inside your charting platform. Detection is instant because the fill happens in the same process, but the copier can only reach accounts that platform can log into, and your machine is the infrastructure: if NT8 isn't running, nothing copies.

Desktop copiers (TradeCopia's classic tier and similar) run as their own program on your computer, talking to broker APIs. More reach than an add-on, but the uptime problem is still yours — which is why most desktop copier vendors also sell hosted VPS tiers: the architecture creates the bill.

Cloud copiers (Trading Floor is one) run the whole pipeline server-side, holding each broker's API connection directly. Your browser is a viewport, not a dependency. The trade-offs run the other way: you're trusting the vendor's infrastructure and security practices, so the vendor owes you visibility — which is where measurement comes in.

Where the milliseconds go

A copy's total delay is a chain: fill detection (websocket push vs polling — polling can add hundreds of ms on its own), decision time (sizing, risk checks, order construction — should be single-digit ms), submission (one API call per follower, ideally in parallel), and finally the broker's own wire — the network round-trip that belongs to the broker, varies by gateway and session, and affects every copier equally.

An honest copier separates those legs, because they have different owners. Our engine stamps every order with its own share and the broker's share, prints both in your Copy Log, and publishes the platform-wide aggregates live at /latency — median engine share is ~42ms on real production copies.

What actually matters when you evaluate one

• Reach: can it connect the accounts you actually hold? (TopstepX and Volumetrica-based accounts don't live inside NT8; add-ons can't reach them.)
• Uptime ownership: whose problem is it at 9:29 ET when your machine decides to update Windows?
• Verifiable latency: per-order numbers you can read, not adjectives.
• Failure behavior: what happens on a partial fill, a rejected order, a dropped connection? Ask. The answer tells you how much production reality the vendor has seen.
• Risk enforcement: a copier multiplies your mistakes across every account — the serious ones ship auto-flatten guardians that act before a firm's drawdown number.

We keep an honest field guide of the copiers we compete with — strengths included — at /best-futures-trade-copier.