Compose strategies with natural language. No code.
Describe what you want in English. EX7’s Strategy Builder emits a validated, executable graph: entry signals, regime filters, position sizing, stops, and exits, wired up and ready to backtest.
Three ways traders use the Strategy Builder.
Translate a thesis into a graph.
English in, executable out.
You have an idea: “mean-revert MNQ off the 9:30 open when VIX is below 18.” Type it in. The Builder generates the entry, the regime filter, the stop, and the exit, fully wired, in seconds. No Python. No TradingView Pine. No fighting an IDE.
Iterate ten ideas a day.
Cheap to compose, expensive to validate.
Composition is fast. Validation is slow. The Builder lets you generate ten variations of a thesis in an afternoon (different regime filters, different stops, different sizing rules), and you only spend Judge runs on the ones that look promising at a glance.
Audited defaults out of the box.
No silent footguns.
Every node ships with parameters that have already been audited against our non-negotiable risk constraints: RTH-only, commission applied, 2-tick slippage on both legs, $500 daily loss cap, three-loss circuit breaker. You can override any of them, but you can’t accidentally ship a strategy that ignores them.
Five questions traders ask.
Do I really not need to write any code?
Correct. You describe what you want in English (“buy MNQ when the 5-minute z-score drops below -2 in low volatility, target 1.5R, stop 1R, only between 9:50 and 15:55 ET”), and the Strategy Builder emits a typed graph with the right nodes wired up. Behind the scenes it compiles to the same execution path as our internal Python strategies, so what you describe is what runs.
What happens if I describe something the system can’t represent?
The validator runs up to three retry attempts with the language model, narrowing the prompt each time. If the third attempt still produces an invalid graph, you get a structured error explaining which node, parameter, or connection broke, not a generic “sorry, try again.” Most failures are vocabulary mismatches (“VWAP” vs. “volume-weighted average”) and resolve on the second pass.
Can I edit the generated graph by hand?
Yes. Every node exposes its parameters as type-safe dropdowns and number inputs. You can tweak the z-score threshold from -2 to -1.8, swap RTH for ETH, change the regime filter, or rip out a node entirely. The graph stays valid as you edit; invalid configurations are blocked at compose time, not at backtest time.
Are there usage limits?
The Scout (free) tier gets 5 generations per month. Operator gets 50, Strategist gets 250, and Principal is uncapped. Each generation includes the full validate-and-retry loop, so you don’t burn quota on language-model hiccups. You can also clone and edit existing graphs without a generation; only fresh natural-language composes count against the quota.
Once I have a graph, how do I know if it’s any good?
Pass it to The Judge. The whole point of separating composition from validation is that you can describe ten ideas in an afternoon, run each through walk-forward + Monte Carlo + DSR, and only paper-trade the ones that PASS. The two tools share a binary format, so the round-trip is one click.
Your idea is ten seconds from a graph.
Join the waitlist. The Scout tier is free: five strategy compositions per month, no credit card.
Start free