Freqtrade covers research, dry-run, and live crypto execution
Freqtrade joins several jobs that are often split across notebooks and scripts. It downloads exchange data, runs backtests, searches strategy parameters, simulates orders in dry-run, and can execute live trades. The same project supplies SQLite persistence, FreqUI, plotting, and optional Telegram commands. FreqAI adds self-training prediction models for users who want that path. The sensible center of the product is still a Python strategy whose entry, exit, pricing, stake, and risk behavior the operator can read.
Our checkout contained 776 files and about 145,360 lines of source in 60.9 MB. That size reflects a trading application with exchange handling, data tools, API services, strategy callbacks, and machine-learning support. Ten CI workflow files, a Dockerfile, a compose file, and a tests directory were present. Those signals match the breadth of the command line, which includes data conversion, backtesting, hyperparameter search, bias analysis, web serving, database conversion, and live trade operations.
Docker starts the service, then strategy work begins
The Docker quick start explicitly says it does not work without further reading and configuration. It creates a user directory and an interactive config, then starts with SampleStrategy. The guide labels that strategy as a demo and tells users to backtest and spend time in dry-run before risking money. Exchange selection, pair lists, pricing, stake rules, and strategy code remain decisions. Pulling an image is the short part of the setup.
A full source install in our 3-CPU, 8 GB Debian sandbox took 144 seconds, added 270 packages, and consumed 6,078 MB. The build needed another 9 seconds. That is heavier than the README's 1 GB minimum disk recommendation for running the bot, though our figure includes the development and test environment rather than a trimmed production image. Use the official stable image if reproducible operations matter, then pin its tag and read the changelog before updating a bot that can place orders.
What happened when we ran it
Our sandbox ran commit e69b3d8 with no credentials or secrets. Installation succeeded in 144 seconds, and the build passed in 9 seconds. Pytest completed successfully in 420 seconds, reporting 4,516 passed, 0 failed, and 20 skipped of 4,516. The dependency audit found 0 known vulnerabilities. Across install, build, and tests, the slowest step was the test suite, which is a reasonable cost for a codebase that can move real funds.
The successful run occupied 6,078 MB across 270 installed packages. It did not establish strategy profitability, exchange compatibility for a particular account, order latency, or live reliability. Those questions need an exchange-specific configuration and market workload. The lab result answers a narrower engineering question: the measured commit installed, built, and passed its available tests in a fresh unprivileged container. For maintainers evaluating changes, 4,516 passing tests are much better evidence than screenshots of a profitable backtest.
Backtests include bias checks, while live callbacks run differently
Freqtrade's lookahead analysis reruns a strategy and compares changed indicators and entries against a baseline. It catches common mistakes such as reading future candles or calculating an average across the full data frame. The documentation also names limits: signals that never trigger cannot be verified, some order options can create false positives, and FreqAI target indicators can be falsely flagged. This is a diagnostic tool, so a clean report does not certify a strategy.
The 420-second suite and 4,516 passing tests support the framework's mechanics, while the strategy remains user code. Backtesting calls many callbacks once per candle; live mode usually calls them once per processing loop, around every 5 seconds by default. The bot basics page warns that this difference can change behavior. Historical pair lists can also hurt reproducibility when they reflect current market membership. A credible workflow includes bias analysis, fixed data, dry-run, and review of live logs.
Exchange credentials and an accurate clock are operational requirements
Dry-run is the right first deployment because live mode adds exchange credentials, account permissions, balances, and irreversible orders. The README requires an accurately synchronized clock for exchange communication. FreqUI is convenient on localhost:8080, though the Docker guide says it lacks built-in HTTPS and should not be exposed directly from a remote server. The suggested access path is an SSH tunnel or VPN. Trade state sits in user_data/tradesv3.sqlite, so backups and one clear owner per database matter.
The repository includes a Dockerfile, compose file, 10 CI workflows, and 270 packages in our installed environment. Operators still need log retention, credential rotation, data backups, and an update policy. Release notes can contain behavior changes: version 2026.8 enabled caching for informative functions by default and removed Bitmart support. Read those notes before pulling a new stable image, restart in dry-run, and confirm that the resolved configuration still matches the intended pairs and order behavior.
September maintenance supports the 2026.8 release
GitHub showed 53,974 stars and 31 combined issues and pull requests when fetched. The repository was pushed on September 3, 2026, three days after release 2026.8. Issue handling was current too: issue 13523 reported that repeated filled-order processing could erase a stored futures funding fee. A maintainer closed it the next day, said the bug was fixed, and added a regression test. That response is more useful evidence of health than the star count alone.
Our run gives Freqtrade an unusually clean engineering result for this category: 4,516 passed, 0 failed, 20 skipped, and 0 known dependency vulnerabilities. Use it if you can write or inspect Python and accept that automation multiplies both a sound strategy and a bad one. Keep the bot on a synchronized Linux host, protect FreqUI, start with dry-run, and test exchange-specific behavior. Buyers seeking stock trading, a no-code profit machine, or permissive licensing should choose another system.