Custom indicators in Python

A LumiBot custom indicator is a small Python class that turns price bars, and any outside data, into numbers you can chart, backtest and trade with. The same class updates live one bar at a time, runs inside a backtest, runs inside a live strategy on your own broker, and can use numpy, pandas, a trained machine-learning model or any web API. LumiBot is free and open source (GPL-3.0) and runs on your own computer with pip install lumibot.

This page answers the main questions in plain words. Each section links to a page with the full detail and tested code.

What a custom indicator can do

Capability

In plain words

Live updates

Updates one bar at a time. A still-forming bar gets a value, then that work is thrown away so it never leaks into history. More: Stream an indicator live from a crypto exchange.

Backtests

The same class runs in a LumiBot backtest and only sees bars that existed at each moment. More: Using an indicator in a backtest.

Live strategies

The same class runs in a live or paper strategy on any broker LumiBot supports. More: Using an indicator in a live strategy.

Streaming without a paid feed

For crypto, a plain Python script can stream public exchange candles through CCXT with no API key. More: Stream an indicator live from a crypto exchange.

Outside data

An indicator can call any web API (for example a funding rate), using only data that existed at each bar’s time. More: Using outside data in an indicator.

Machine learning

Plain Python, so numpy, scikit-learn or a model file you trained all work. More: Indicator examples.

Settings and outputs

Declares settings people can change, the lines it draws, colors and guide lines. More: Indicator settings and outputs.

Chart placement

Draws on the price chart or in its own pane. More: On the price chart or in its own pane.

TradingView numbers

Built-in sma, ema, rsi, macd, bbands, atr, stoch and vwap follow Pine Script v6 formulas. More: Matching TradingView’s numbers.

How an indicator works in one picture

bars (history, then live)          your indicator              you
-------------------------   --->   update(bar, closed)   --->  values per bar
open high low close volume          keeps a small state        {"rsi": 61.2}

  • closed=True means the bar is finished. The indicator commits it.

  • closed=False means the bar is still forming (a live tick). The indicator gives a value for it but throws its work away afterwards, so the forming bar never leaks into history. TradingView calls this “rolling back”.

Because each new bar costs one update, a live chart does not recompute all of history on every tick. To write one, start with Build your first custom indicator (about five minutes).

How LumiBot compares, in one table

A summary of what an indicator author can do in each tool. The full tables, plain-words notes and every source are on LumiBot indicators compared (checked October 25, 2026).

Product

Language

Outside web data

Same code places live broker orders

LumiBot

Python

Yes, inside the indicator

Yes, inside a strategy on your broker

TradingView

Pine Script

No

Through alerts and webhooks to other services

QuantConnect LEAN

Python or C#

Yes, in the algorithm (limits apply in their cloud)

Yes, LEAN algorithms

NinjaTrader 8

C# (NinjaScript)

Possible with .NET classes, not supported by NinjaTrader

Yes, NinjaScript strategies

MetaTrader 5

MQL5

No; web requests are blocked in indicators

Yes, Expert Advisors

LuxAlgo Vela + PineTS

Pine run in TypeScript

No; the chart feeds the data

No (paper trading)

OpenMarket (wrun)

TypeScript to WebAssembly

Built-in feeds only

No

Frequently asked questions

What is a custom indicator in LumiBot?

A Python class that reads price bars one at a time and returns values, such as an RSI line or a channel. It declares its settings and outputs, so a chart knows how to draw it, and it works in backtests and live strategies without changes.

Is LumiBot free?

Yes. LumiBot is free and open source under GPL-3.0. Install it with pip install lumibot. Nothing on these pages needs an account or a hosted service.

Can an indicator use machine learning?

Yes. It is plain Python, so numpy, scikit-learn, LightGBM or a model file you trained all work. Indicator examples includes a small numpy classifier.

Can an indicator use data from outside the chart?

Yes. It can call any web API. Using outside data in an indicator shows a funding rate from a public exchange, done without look-ahead in backtests.

Can I get live indicator values without a paid data service?

For crypto, yes. Stream an indicator live from a crypto exchange streams public candles from an exchange through CCXT with no API key. Stocks need a data source or broker that provides them.

Will my indicator repaint?

Closed-bar values never change. A forming bar’s value can change until the bar closes, which is why LumiBot labels it closed=False and never saves it into history.

Does an indicator make money?

No tool can promise that. Indicators describe the past. Test any idea in a backtest and on a paper account first.

All indicator pages