LEAN

What it is

The event-driven engine QuantConnect's cloud runs, published on its own under Apache-2.0. The core is C#; algorithms are C# or Python 3.11. One algorithm class runs the backtest, the paper session and the live deployment, so research code and trading code cannot drift apart.

This card is the engine you host. The browser IDE, cloud compute and mounted data are QuantConnect, sold per seat and per node. You come here to keep the execution, the data and the broker login on your own hardware.

As of 7 October 2026 the last commit on master was 6 October, which also produced build tag 18167, NuGet QuantConnect.Lean.Engine 2.5.18164 and a new Docker image. About 21,900 stars, 238 open issues and 22 open pull requests; not archived. Builds are numbered tags, three on 6 October alone, so the version is whichever build you pulled. The GitHub releases tab is stuck at v2.4.0.1 of August 2017 and tells you nothing: versioning moved to continuous builds years ago.

Pricing

Nothing in the public repository is held back. The money is around it:

• Built from source or run from the Docker image: free, no account.
• The LEAN CLI (pip install lean, 1.0.229 of 28 August 2026) is Apache-2.0 too, but its documentation says it requires a paid QuantConnect organization — from a $10/month Researcher seat ($96/year).
• Private Cloud Hosts, $4,800/year, runs a cluster of LEAN servers on your own machines.
• LEAN Enterprise, a proprietary build with speed improvements not pushed to the open source, runs on-premises only on the Institution tier.

QuantConnect's own pages disagree on local use: the CLI docs say any paid tier runs LEAN locally, while the pricing table marks on-premises backtesting and live trading as "Cloud" below Institution. Settle that before buying a seat for local work.

Data & coverage

LEAN ships no data you could trade on. The samples cover daily bars for 21 US tickers and minute bars for 10, mostly one week of October 2013. Data lives on disk as zipped CSV under /data/securityType/market/resolution/ticker/; you fill it from a third-party feed or buy QuantConnect's AlgoSeek-sourced files. On a Researcher seat:

• the US Equity Security Master (splits, dividends, ticker changes) is $600/year — $1,800 at Institution — and every equity download depends on it;
• bulk US equity history is $2,136/year for daily or hour bars, $11,760 for minute and $16,800 for tick, then $600/year for updates. Minute is 500 GB on disk, tick 1.5 TB;
• by ticker, $1 per security for daily bars and $0.05 per security-day for each minute file;
• bulk equity options at minute resolution total $46,320 in year one with prerequisites, and 6 TB.

Third-party feeds are plugins: Massive (still called Polygon in the docs) for equities and options, ThetaData, IQFeed on Windows only, and Databento for US futures only. The Polygon plugin serves unadjusted prices and the docs send you to the $600 Security Master for adjustments; without it a split reads as a crash.

Integrations

For a US stock or options account, five brokerage plugins have public source: Interactive Brokers, Tradier, TradeStation, Alpaca and tastytrade. Charles Schwab, Webull, Public, Clear Street and Eze are on the CLI's live-trading list, but their code is not public — the docs link a Schwab repository that returns 404 — so they arrive as prebuilt modules for a logged-in, paid account.

With Interactive Brokers, LEAN by default starts IB Gateway inside its container, logs in, pushes the IB Key prompt to your phone and restarts the gateway weekly at a time you set; or you run your own gateway and pass its host and port.

The default brokerage model charges equities, options and futures at IB's schedule — $0.005 a share, $1 minimum, capped at 0.5% of trade value — with zero slippage and a margin account; a cash account settles stock at T+1. Results are JSON, plus an HTML or PDF tearsheet from lean report.

Limitations

• No alerts from a self-hosted live algorithm. The engine refuses email, SMS, webhook, Telegram and FTP notifications and logs that sending is not implemented. If it stops, nothing tells you.
• The config lists brokers you cannot use. Launcher/config.json keeps live environments for FXCM, FTX, FTX US and Exante, whose plugins are archived.
• Bought data stays in LEAN. The CLI data agreement allows individual or internal use only and says the files may not be converted for other applications — a $12,360 minute-bar purchase does not move to another backtester.
• The docs lag the code. The getting-started page asks for .NET 9 on macOS and .NET 6 on Linux; the solution targets .NET 10. lean.io advertises T+3 settlement; the code settles stock at T+1.
• It is heavy. Docker Hub reports the amd64 image at 14 GB and arm64 at 5 GB.

Alternatives

NautilusTrader is the closest match: backtest to live unchanged, Interactive Brokers and Databento for equities, a Rust core, LGPL. QuantRocket is the other self-hosted Docker platform wired to Interactive Brokers, built on Zipline and Moonshot. If the backtest never becomes a live order, Backtrader or Zipline-reloaded are a pip install rather than a .NET solution, and ib_async gives you the IB connection without an engine.