Open source crypto trading bot: licenses, status, checks

Short answer: An open source crypto trading bot is software whose code you can read, run and change under its license, while hosting, keys and results stay your responsibility. As read on 5 November 2026, Freqtrade and OctoBot are GPL-3.0, Hummingbot, LEAN and Superalgos are Apache-2.0, Jesse is MIT and NautilusTrader is LGPL-3.0. Before running any repository, read it, run it in a container with no real keys, and pin the version; trading with any of them can lose money.

This is a reference, not a ranking: alphabetical, no star counts, no strategy, preset or performance figure. The components a production system needs are in trading bot development, and how three of these projects are built is in trading bot architecture.

Which venues you may trade on depends on your jurisdiction and each platform's terms, and no exchange is recommended here. Article 99/A of Capital Markets Law No. 6362 treats activity by foreign crypto-asset platforms directed at Turkish residents as unauthorized.

Open source trading bots on GitHub: seven projects compared by fact

All seven repositories showed commits in 2026 when read, but the last formal GitHub releases of the six that publish them range from 30 October 2026 back to 2017.

Read the release and commit columns together. LEAN's last GitHub Release is old while its numbered tags and commits are current, and its README points to its CLI, Docker Hub and NuGet. These dates will age; recheck them.

Python trading bot framework or ready-made bot: what differs

Which project wins has no useful answer; compare scope, runtime and conditions, as the projects state them.

Some also say where they can fall short. Freqtrade's README says to use it at your own risk and to start in dry-run. NautilusTrader's says live execution brings behavior "a simulation may not reproduce". A library such as CCXT is a different thing again, a connector rather than a bot; see the CCXT Python guide.

What the license means if you build a product on it

Running a bot for yourself is allowed under all four licenses; the differences start when you hand copies to someone else. This is a reading of the texts, not legal advice.

All four disclaim warranty, and a code license does not settle other rights. OKX TR's API agreement (opens in a new tab), for example, restricts SaaS, signal and managed services in clause 9.3 and redistribution of market data in clause 9.4 without written approval. Running software for other people's accounts raises separate questions, covered in are trading bots legal.

How to check a trading bot from GitHub before running it

Treat any repository as hostile until you have read it, because a polished page proves nothing about the code.

Kaspersky's GitVenom report (opens in a new tab) of 24 February 2025 describes hundreds of fake GitHub repositories with well-written READMEs, tags and commit counts inflated by a timestamp file. Python samples hid code after about 2,000 tab characters, JavaScript samples decoded a Base64 script, and C-family projects ran a script through a Visual Studio PreBuildEvent. Payloads stole credentials and wallet data, installed backdoors and swapped copied wallet addresses. Its named examples are an Instagram tool, a Telegram Bitcoin wallet bot and a game hack, not trading bots; a bot that asks for exchange keys looks like the same kind of target, though that is an inference.

1. Find the real repository from the project's own site or package page, not from a search result, and check the owner.
2. Scan and read. The script below looks for those three tricks.
3. Run it locked down, with no network at first, then only the hosts it needs.
4. Use no real keys at first, then a key with read and trade permission and no withdrawal right; see API key security.
5. Pin what you tested: a tag and hash, or an image digest.

import pathlib, re, sys

CHECKS = {
    "hidden off screen": re.compile(r"(?:^|[^ \t])[ \t]{200,}\S"),
    "decode and run": re.compile(r"^(?=.*(exec|eval|Function)\s*\()(?=.*(b64decode|base64|atob|decompress|[Ff]ernet))"),
    "runs at build or install": re.compile(r'PreBuildEvent|PostBuildEvent|"((pre|post)?install|prepare)"\s*:'),
}
SUFFIXES = {".py", ".js", ".ts", ".csproj", ".vcxproj", ".json", ".toml", ".sh", ".bat", ".ps1"}

root = pathlib.Path(sys.argv[1])
if not root.is_dir():
    sys.exit(f"not a folder: {root}")
hits = 0
for path in sorted(root.rglob("*")):
    if path.is_file() and path.suffix.lower() in SUFFIXES:
        for n, line in enumerate(path.read_text(errors="ignore").splitlines(), 1):
            for name, pattern in CHECKS.items():
                if pattern.search(line):
                    hits += 1
                    print(f"{path.relative_to(root)}:{n}: {name}: {' '.join(line.split())[:40]}")
print(hits, "hits; a hit means read that line, and no hit proves nothing")

On a synthetic fake repository built on 5 November 2026:

bot.csproj:1: runs at build or install: <Project><PropertyGroup><PreBuildEvent>r
index.js:2: decode and run: eval(atob('Y29uc29sZS5sb2coImhpIik='));
main.py:3: hidden off screen: time.sleep(0) ; import base64; exec(base
main.py:3: decode and run: time.sleep(0) ; import base64; exec(base
package.json:1: runs at build or install: {"name":"x","scripts":{"postinstall":"no
5 hits; a hit means read that line, and no hit proves nothing

The same script found 0 hits on a clean sample, the CCXT 4.5.92 package and Freqtrade's 2026.10 source; a quiet scan is not a safety verdict on any project.

Run it in a container and read the network calls

A container with no network and a read-only root limits what a surprise can do, and a Python audit hook shows what the code tries.

import runpy, sys

WATCHED = {"socket.connect", "socket.getaddrinfo", "subprocess.Popen", "os.system"}

def hook(event, args):
    dynamic = event == "compile" and args[1] in (None, "<string>")
    if event in WATCHED or dynamic:
        print("AUDIT", event, repr(args)[:60], file=sys.stderr)

sys.addaudithook(hook)
runpy.run_path(sys.argv[1], run_name="__main__")

docker run --rm --network none --read-only --cap-drop ALL \
  --security-opt no-new-privileges --user 65534 --pids-limit 64 --memory 256m \
  -v "$PWD:/repo:ro" python:3.12-slim python -I /repo/watch.py /repo/fake/main.py

Run on 5 November 2026 against the fake repository, whose hidden line ran a harmless payload aimed at a reserved documentation address (standard error only; three standard-library compile lines and the traceback body left out):

AUDIT compile (b'import socket, subprocess\nsubprocess.run(["echo", "paylo
AUDIT subprocess.Popen ('echo', ['echo', 'payload ran'], None, None)
AUDIT socket.getaddrinfo ('192.0.2.1', 443, 0, 1, 0)
AUDIT socket.connect (<socket.socket fd=3, family=2, type=1, proto=6, laddr=('0.0
OSError: [Errno 101] Network is unreachable

The hook exposed the decoded code, the subprocess and the connection, and --network none stopped the connection. Python's documentation warns that audit hooks are not suitable as a sandbox and that malicious code can bypass them, so the hook is for looking and the container is the barrier. A container is not a complete boundary either: mount nothing you care about, and never put keys in the image.

Pin versions: pip may not install what the README describes

Pin an exact version and verify it, because pip quietly falls back to an older release when your Python is too old for the current one.

--only-binary :all: matters here, since building a source package can run its code. For pinning, use pip install --require-hashes with a lock file, or an image digest. A Freqtrade image pulled earlier to this machine, run with --network none and --version, reported Freqtrade 2026.10 with CCXT 4.5.90, digest starting sha256:bd52747f, so a bot's bundled exchange library can lag the current release (CCXT 4.5.92 that day). That was the only framework run here, and nothing traded.

If you would rather have someone else read a fork or your changes, trading bot code review covers that. Whichever project you pick, tests show how the software behaves, not future results, and trades can lose money.

Frequently asked questions

Is there a free open source crypto trading bot?

All seven publish their code at no charge. You still pay for the server and your time, and some features have conditions, such as Jesse's plugin key for paper and live modes.

Which open source trading bot should I choose?

It depends on your venues, language and license needs, and on whether you want a ready bot or a framework.

Can I sell a bot built on Freqtrade?

The GPL allows charging, but recipients get the source under the GPL and may share it. Have a lawyer check your plan, and read the venue's API terms too, since they apply separately.

Does a busy GitHub page mean a repository is safe?

No. Kaspersky's GitVenom report describes fake repositories with polished READMEs and inflated commit counts. Check the owner, read the code and run it locked down.

Sources

• Freqtrade repository and README (GPL-3.0, supported exchanges) (opens in a new tab)
• Hummingbot repository and README (Apache-2.0) (opens in a new tab)
• Jesse repository and README (MIT) (opens in a new tab)
• Jesse documentation: Live Trading (plugin and license key) (opens in a new tab)
• QuantConnect LEAN repository and README (Apache-2.0) (opens in a new tab)
• LEAN CLI documentation: QuantConnect Paper Trading (opens in a new tab)
• NautilusTrader repository and README (LGPL-3.0) (opens in a new tab)
• OctoBot repository and README (GPL-3.0) (opens in a new tab)
• Superalgos repository and README (Apache-2.0) (opens in a new tab)
• Kaspersky Securelist: GitVenom campaign, 24 February 2025 (opens in a new tab)
• GNU General Public License v3.0, as shipped in Freqtrade's LICENSE file (opens in a new tab)
• GNU Lesser General Public License v3.0, as shipped in NautilusTrader's LICENSE file (opens in a new tab)
• Apache License, Version 2.0 (opens in a new tab)
• Python documentation: sys.addaudithook (opens in a new tab)
• Capital Markets Law No. 6362, consolidated text, Article 99/A (Turkish) (opens in a new tab)

This guide is general information about software and markets, not investment, legal or tax advice. Automated trading can lose money, past or backtested performance does not guarantee future results, and the rules that apply depend on your jurisdiction and your exchange's terms. Sources, checks and corrections: see the editorial policy.