What is Stochastly?

Stochastly is a desktop application for Windows and macOS in which traders and quantitative researchers build a trading strategy as a node graph, backtest it on their own data and measure its risk of overfitting before exporting it to a trading platform.

The name refers to this application, published at stochastly.com by the publisher named in the legal notice and described on the about page. It is unrelated to the stochastic oscillator, a momentum indicator of technical analysis, and to software sold under similar names.

Key facts

Stated as of . Each line links the page that carries the fact in full.

editionsStarter, Atelier and Portfolio, planned; none is on sale, and the site sells one thing, an optional refill of assistant credits (join the early access)
AI add-onAssistant + Agents, an optional subscription on any plan, 500 AI credits included each month
accessno free plan; early access: free for life on one computer, in exchange for written feedback; an account and five short questions before the download
platformdesktop application for Windows 10 or 11, x64, per-user installer downloaded from your account (install guide)
macOSmacOS 13.1 or later on Apple silicon; a .dmg from your account, signed with a Developer ID and notarised by Apple on 2026-10-16 (install guide)
Linuxno build, and no Linux direction on the roadmap
version0.2.4, alpha; 25 dated entries from 2026-08-12, the first build, to 2026-10-18 (version history)
live tradingthrough an exported file run where you trade (Pine Script on TradingView, MQL5 on MetaTrader, Python in your own setup); direct broker execution from the app is planned once the applicable regulation is in place (roadmap)
where it runson your computer: price files, strategies and results stay in the local Vault; the licence check sends the licence token, the device secret and a hashed machine fingerprint, and the optional hosted assistant sends its conversations to the Stochastly relay and on to Nebius B.V. in the Netherlands (network traffic · privacy)
publisherMahé Chadeffaud, sole proprietor trading as Stochastly, SIREN 109 270 363, Annecy, France (legal notice); founder and co-founder on the credits page

The short answer

categoryquantitative research and backtesting software
datayour own price files, read locally
forsystematic traders, prop-firm candidates, quantitative researchers and small research teams

What it does, step by step

1. Build the strategy

A strategy is a graph of nodes, drawn in a visual editor, with a catalogue of 667 nodes across 28 families: 492 validated by executable checks, 175 marked roadmap. A Python editor sits next to it for custom nodes. Node reference.

2. Backtest it without leaked data

Before a run, a causality gate truncates the bars and replays the graph; a computation whose past values move when future bars are removed is refused by name. The backtest then runs locally with trading costs in basis points and commissions. The causality gate.

3. Measure the overfitting

Each run reports walk-forward results, the deflated Sharpe ratio, the probability of backtest overfitting (PBO), Romano-Wolf StepM adjusted p-values and combinatorial purged cross-validation folds. These statistics answer one question: how much of the backtest survives once the number of trials is counted. Measurement protocol.

4. Take the strategy out

A validated strategy compiles to 6 targets from one representation: Python, C++, Rust, Pine Script and MQL5, each file stating what it reproduces. From Atelier, the Paper Editor produces a typeset research paper in PDF that documents the run and its verdict. Starter has no Paper Editor access. Exports · Research papers.

5. Drive it from an agent

An MCP server with 35 tools lets an AI client such as Claude Desktop compose a graph, launch a run and read the verdict. An optional assistant inside the app does the same from a conversation. The assistant is hosted and is the one feature that sends content off your machine; with it switched off, strategies, data and results stay on the computer. MCP server · Assistant.

MCP launcher source on GitHubstochastly-mcp · MIT

Three typical uses

A trader with an idea

The idea becomes a graph, the backtest runs on the trader's own history, and the trader reads the overfitting statistics before taking it further. A strategy that holds up is exported to Pine Script for TradingView or MQL5 for MetaTrader and traded there.

A prop-firm candidate

A challenge has a profit target and a loss limit, so the relevant figure is the probability of reaching the target first. Stochastly computes that pass probability from the strategy's own returns. Prop-firm evaluations.

A researcher writing up a result

Several strategies are compared in one place, the multiple-testing correction is applied across the compared strategies, and the run exports as a paper with its disclosures. Statistical methods.

Where it sits among trading tools

cloud platformsStochastly runs on your machine, on data you already hold
Python librariesStochastly adds a visual editor and reports overfitting statistics for each run
strategy generatorsStochastly counts the trials behind a result and deflates the Sharpe ratio by them
charting platformsStochastly exports the validated strategy to Pine Script and MQL5 for them

Frequently asked questions

Is Stochastly available on macOS or Linux?

On macOS, yes: the account offers a .dmg for macOS 13.1 or later on Apple silicon, signed with a Developer ID and notarised by Apple. Windows gets an installer for Windows 10 or 11 on x64. Intel Macs are not supported. There is no Linux build and no Linux direction on the roadmap.

Can I trade live with Stochastly?

Yes, through an export: a validated strategy compiles to a file that you run where you trade, Pine Script on TradingView, MQL5 on MetaTrader or Python in your own setup. Direct broker execution from the app is planned once the applicable regulation is in place, and its state is tracked on the roadmap. Exports.

Where does my data go?

Price files, strategies and results are read and written on your computer, in the local Vault, and a backtest run emits no telemetry unless you opt in on the sign-in screen to one of the two data-sharing purposes; both start unticked. The requests that leave the machine are listed host by host on the network traffic page, generated from the application source. Among them, the licence check sends the licence token, the device secret and a hashed machine fingerprint, and the optional assistant sends its conversation, as described in the answer below. What the account and the website store is set out in the privacy policy.

Can I use Stochastly today?

Yes, through the early access: an account, five short questions, then the download. It is free for life on one computer, in exchange for written feedback. No edition is on sale yet. Join the early access.

Does the AI assistant send my strategies?

Parts of them, while you use it. The assistant is hosted: its conversation, which can include the graph you have composed, the file open in the editor and figures it quotes from a run you ask it to read, goes to the Stochastly relay and on to Nebius B.V., in the Netherlands, which runs the model. Bar files and result files are not uploaded, and Nebius states that it does not use this content to train any model. Switched off under Settings, the assistant sends nothing to the relay. Assistant · privacy policy.

How mature is Stochastly, and what does alpha mean?

The desktop app is in alpha at version 0.2.4, and the version history dates 25 entries from its first build on 2026-08-12 to 2026-10-18. Settled in this alpha: the 492 catalogue nodes validated by executable checks, one release number shared by the app, the installer and the agent interface, and exported files that state what they reproduce. Still moving: the 175 nodes marked roadmap, parts of the interface, an installer without a code signature yet, and the undated directions on the roadmap.

How is Stochastly different from QuantConnect or TradingView?

QuantConnect runs backtests in its cloud on data it hosts; its free plan includes unlimited backtesting, and live trading through a brokerage comes with its paid tiers (QuantConnect pricing, read 2026-10-03). TradingView runs Pine Script strategies on the bars of its charts and shows a strategy report in a tab below the chart (Pine Script strategies, read 2026-10-03). Stochastly runs on your computer against your own files, reports the deflated Sharpe ratio and the probability of backtest overfitting for each run, and exports a validated strategy to Pine Script for TradingView or MQL5 for MetaTrader. Comparisons · Stochastly and QuantConnect.

What comes next

Direct broker execution and managed-account features are planned. They open once the applicable financial regulation is in place; until then, live trading goes through the exported files on the platform of your broker. Roadmap · current version.

Join the early access Install guide