Why use Anthid instead of building your own execution layer

Anthid is an order and execution management layer delivered as an API. Your systems submit an instruction. Anthid records it, checks it against the limits you set, routes it to your broker, streams back what happened, and keeps the whole record. Broker connectivity is one capability of that layer, not the layer itself. Every developer who automates a strategy eventually decides whether to build that layer or run on one. Building it feels productive at first. The broker’s API docs look manageable, the first test order goes through, and it seems like you saved money. Then reality shows up.

​

What rolling your own actually costs

The broker client is the visible part of the work and the smallest. What sits around it is the layer, and the layer is a permanent obligation. Orders get rejected for reasons the docs never mention, and you need somewhere to record that the instruction existed before the rejection came back. A websocket drops at 2am and your fill events quietly vanish, so you need a record to reconcile against. Positions drift out of sync with the broker and you lose a morning working out which side is right. A retry double-sends an order because nothing made the submission idempotent. A risk check you meant to write properly is still a comment. Then you add a second broker and discover that everything above the client took the first client’s shape: its order formats, its error codes, its idea of what a “filled” event even looks like. Here is the uncomfortable part. None of that work makes your strategy better. It keeps the lights on, and that is all it does. The hours you spend on the layer are hours you are not spending on the thing that actually makes money.

​

What Anthid does instead

Anthid runs that layer as a service, and it is built around one rule: the instruction is recorded before anyone is told it succeeded. You describe the outcome you want as an intent, in Anthid’s own terms. The intent is written to an append-only log in the same transaction that queues it for the broker, so no order can reach a venue without a record. A retry under the same Idempotency-Key is answered with the intent the first request made rather than placing a second order; see Request correlation and retries. A replace or a cancel is appended as a new sequenced action rather than overwriting the original, so the full history of what was instructed is always reconstructible. And the platform reports two things separately: how far it got in handing the order to the broker, and what the broker then did with it. Sent is never mistaken for filled. Controls run before the order ever leaves the platform. Trading windows and maximum order size are enforced at submission, so an oversized order, or one sent outside approved hours, is stopped before it reaches the broker rather than discovered after the damage is done. Organization controls act as a kill switch across every connected account, which is the fastest way to stop trading during an incident, and every refusal is recorded with the same attribution as an accepted order. Every order, fill, and position update arrives in one normalized stream and lands in a durable ledger that is kept for six years, so when you ask “what actually happened to that order,” there is a real answer. The broker sits behind all of that. An intent is translated into each broker’s dialect at routing time, so the instruction, the controls, the stream, and the record are the same whichever connected account is on the other end. Lightspeed is connected today, in beta, more are on the roadmap, and adding one is a config change rather than a rewrite.

​

The numbers you would otherwise own

Anthid runs at 99.99% uptime with median order acknowledgement of 42ms. Those figures are easy to read and very hard to earn. If you build your own layer, you own them yourself, along with the monitoring, the failover, and the 3am pages when something slips. Most small teams never get close, and the gap shows up at the worst possible moment, which in trading means an open position you cannot see.

​

When building your own still makes sense

To be fair, custom infrastructure has its place. If you are a large firm with a dedicated platform team, colocated servers, and latency requirements measured in microseconds, you probably should own the whole stack. For a solo developer, a small quant team, or a startup shipping its first strategies, the math points the other way. You are not really choosing between free and paid. You are choosing between paying with months of engineering time or paying for a service that already solved the problem.

​

Try it before you commit

Anthid has a free tier with paper trading, no credit card and no sales call required. Paper accounts run against the same API surface as live, so you can point your strategy at the API, run it against paper money, and see how it feels in an afternoon. Connecting a live broker account moves you to a Starter or Professional plan, but nothing about your integration changes when it does. If the goal is to ship a live strategy this week instead of this quarter, that is the whole argument in one sentence.

​

More articles