AI generation

Overview

What the AI can and can't do, the tools it has, and the contract between you, the model, and your indicator.

StrategyTune lets you describe an indicator in plain English and have a model write the AssemblyScript for you. The model edits the same draft you'd edit by hand, then compiles and (if you ask) installs it on the chart. Once saved, an AI-generated indicator behaves like any other custom indicator — same code, same compile path, same chart.

How the chat works

Open the indicator editor and the chat panel. You describe what you want in plain language; the model writes the indicator and (if you ask) installs it on the chart. A typical exchange:

  1. You: "Build me a Bollinger Bands indicator with a 20-bar window and 2σ multiplier."
  2. The model writes the indicator's settings and code, compiles it, and adds it to your chart.
  3. The bands draw on the chart and you can run the replay.

What the AI can do

The AI is scoped to one indicator at a time. Within that scope it can:

  • Read your indicator's current settings and code.
  • Write new settings and code, or modify what's there.
  • Compile the indicator — that's where AssemblyScript turns into something the chart can run.
  • Add the compiled indicator to your chart.
  • Tell you what's wrong if compilation fails, and try again.

Edits live as a draft until compile succeeds. Nothing the AI writes affects your chart until the indicator compiles cleanly. If compilation fails, the previous working version stays on the chart — you don't get stuck with a broken indicator.

What you have to do

  • Sign in. AI features require an account. Anonymous users get the rest of the app, but not chat.
  • Have credits. AI calls draw from your subscription's monthly credits first, then from any purchased top-ups. See Credits.
  • Pick a model. Three are available — see Models.

What you don't have to do

  • Learn AssemblyScript. The model is briefed in detail on the indicator API and the rules of the language.
  • Manage prompts. There's no prompt-engineering layer to configure — just describe what you want.
  • Pick a coding pattern. The model decides whether your indicator should recompute from history each bar or carry state forward.

What it can't do

  • Run your indicator. The model writes and installs; you start the replay yourself when you're ready to test it.
  • Touch other indicators. Each chat is scoped to one indicator.
  • Change your account, billing, or chart layout.
  • Make claims about backtest performance — it writes the indicator, it can't promise edge.

For the API surface the model is writing against, see Indicator API.

Something missing or wrong? Email support@strategytune.com.