The agent

Most automated trading sits at one of two extremes. On one end: rigid bots that execute static rules with no judgment. Every condition is a number you specify in advance, and any market state that doesn't match the rule precisely is invisible. On the other end: opaque "AI" black boxes that act on logic you can't see or audit, and that you have to trust because you have no other option.

Engine's agent is neither. It reads the market continuously, interprets the strategy you wrote the way a thoughtful trader would interpret it, and trades inside the constraints you set. Every decision it makes is traced, in plain English, in your decision log. The agent is what we wanted to use ourselves and couldn't find anywhere else.

The rest of this page describes what the agent is capable of. It does not describe how the agent is built. That's our problem to solve, and we keep iterating on it.

It reads the market holistically

The agent doesn't act on one indicator at a time. It considers funding, liquidity, order flow, basis, regime, and cross-market correlation together, the way an experienced discretionary trader would weigh them. It does this for every market in your universe, every tick, around the clock.

This is the part that an "if RSI < 30 then buy" bot cannot do. A static rule sees one number and acts on it. The agent sees a market state and interprets it.

It understands strategy bundles

Your strategy is a bundle: STRATEGY.md for the trading playbook, strategy.yaml for hard runtime boundaries, and optional tools.yaml entries for custom context the agent can call by name. The agent interprets the playbook, while Engine enforces the config.

You can describe market conditions no off-the-shelf bot framework supports: regime shifts, calendar events, unusual flow, narrative-driven setups, cross-exchange divergences. The agent will operationalize them. You aren't binding to a parameter; you're binding to a description.

Most products in this space make you pick from a list of supported indicators. Engine doesn't have a list. Anything you can describe precisely enough that another trader would understand you, the agent can act on.

It adapts to changing regimes

When the market shifts (funding flips, vol expands, liquidity moves), the agent's interpretation of your rules updates with the conditions. If you wrote "fade extreme funding," the threshold for extreme on a Wednesday in a calm market isn't the same as the threshold for extreme during a Sunday liquidation cascade. You don't have to manually adjust it. The rules you wrote keep meaning what you meant by them.

This is the practical difference between a strategy that works for two weeks and a strategy that works for a year.

It explains every decision

Every trade comes with a paragraph: what the agent saw, what rule it matched, why it sized the way it did, where the stop is, where the take-profit is, and what data points pushed it from "maybe" to "yes." If a position confuses you, you can ask the agent why, in chat, in plain English, and get the underlying numbers back.

Other "AI traders" don't show their work because there isn't any work to show. The agent's transparency isn't a feature we bolted on; it's the feature.

It improves between sessions, without changing your rules

The agent's market reading and judgment get sharper as we ship updates to it. You don't have to do anything to get the upgrade. Crucially, your strategy bundle doesn't change underneath you. The contract is: your bundle is the source of truth for what the agent does; the agent is the engine for how. We keep making the how better. The what stays exactly the way you wrote it.

What this means in practice

You write a strategy that describes how you want to trade. The agent operationalizes that description, every tick, on every market in your universe, with full transparency about why it's doing what it's doing. When the market shifts, the agent adapts; when we ship an update, the agent gets sharper; when something surprises you, you can ask why and get a real answer.

There isn't another agent like this in the market. Most products in this space are one of the two extremes we opened with. Engine is what we wanted to use ourselves.