Freqtrade & FreqAI Review

This Freqtrade review is for readers who have heard that most "AI trading bots" are rule-based automation with a marketing layer, and want to see what the real thing looks like. Freqtrade is the answer most developers give. It is a mature open-source bot written in Python, and its FreqAI module does what most vendors only claim: it trains machine-learning models on market data and keeps retraining them as conditions change.

That also makes it the least beginner-friendly tool in our best AI crypto trading bots ranking, where it holds the top score for real AI and security. Below is what FreqAI actually does, what running it costs, where people go wrong, and who should not bother.

Freqtrade review at a glance

What Freqtrade is, and what it is not

Freqtrade is a trading framework, not a trading service. You download it, configure it with your exchange keys and a strategy, and run it on hardware you control. A strategy is a Python class that calculates indicators and decides when to enter and exit. Freqtrade handles everything around that: fetching candles, managing orders, tracking open trades, applying stop-losses and reporting results.

The project has been running since May 2017, which makes it older than most commercial bot platforms. It is not-for-profit: there is no company selling you a premium plan, no token, and no signal marketplace. That matters for trust. Nobody involved has a commercial reason to inflate returns, and the documentation never promises any.

What it is not: a plug-and-play AI that finds profitable trades for you. Out of the box you get example strategies meant to show how the code works, not to make money. Everything that might give you an edge, from feature choice to risk rules, is your job.

How FreqAI works: the AI is real

FreqAI is the reason Freqtrade carries our "Real ML" label. It is a module that automates the tedious parts of applying machine learning to price data. You define features (for example RSI, volume changes or returns over several periods, on several timeframes and correlated pairs), define a target (what the model should predict, such as the price change over the next 24 candles), and pick a model. FreqAI then builds the dataset, trains the model, makes predictions on new candles, and passes those predictions to your strategy as extra columns.

Supported models

The documentation covers gradient-boosting models such as LightGBM and CatBoost, PyTorch neural networks, convolutional network examples, and reinforcement-learning agents that learn a trading policy by trial and error in a simulated environment. You can also write your own model class.

Adaptive retraining

Markets drift, so a model trained once goes stale. In dry-run and live mode FreqAI retrains periodically on a sliding window of recent data (set by train_period_days) while the bot keeps trading. In backtesting it simulates that same behaviour by retraining every backtest_period_days, so your backtest reflects a model that only ever saw the past. That design choice is what separates FreqAI from the many scripts that train one model on all available history and then "test" it on data it has already seen.

A minimal FreqAI setup: config and first run

A short FreqAI tutorial helps show the shape of the work. FreqAI is switched on with a freqai block inside your normal Freqtrade config file. The example below is adapted from the official configuration docs, which stress that the values are illustrations, not recommendations.

"freqai": {
  "enabled": true,
  "purge_old_models": 2,
  "train_period_days": 30,
  "backtest_period_days": 7,
  "identifier": "unique-id",
  "feature_parameters": {
    "include_timeframes": ["5m", "15m", "4h"],
    "include_corr_pairlist": ["ETH/USD", "LINK/USD", "BNB/USD"],
    "label_period_candles": 24,
    "include_shifted_candles": 2,
    "indicator_periods_candles": [10, 20]
  },
  "data_split_parameters": {
    "test_size": 0.25
  }
}

Read it line by line: train on the last 30 days, retrain every 7 days in backtests, build features on three timeframes and three correlated pairs, predict 24 candles ahead, add two shifted copies of each candle as extra features, and hold back 25% of the data for testing. Then you start the bot with a strategy that uses FreqAI and a model name:

freqtrade trade --config config_examples/config_freqai.example.json \
  --strategy FreqaiExampleStrategy \
  --freqaimodel LightGBMRegressor \
  --strategy-path freqtrade/templates

Swap trade for backtesting to run the same setup on history first. Two warnings from experience. First, more timeframes and correlated pairs multiply the feature count fast, along with training time and memory. Second, the example strategy exists to demonstrate the mechanics. If you want a guided build from data to paper trading, follow our Python AI bot tutorial before touching real funds.

Core features beyond the AI

Even without FreqAI, Freqtrade is one of the most complete bot frameworks around. The parts you will use most:

Installation is via Docker (the simplest path) or directly with Python 3.11 or newer. Configuration is JSON, strategies are Python files, and all of it lives in a folder you can put under version control, which is a quiet advantage over hosted platforms where your settings vanish if the vendor changes products.

Supported exchanges: spot and futures

Freqtrade connects to exchanges through the CCXT library, but only some venues are officially tested. As of October 2026 the docs list:

Check your own exchange's availability in your country before you start. Freqtrade will happily connect to a venue you are not allowed to use, and that is your problem, not the bot's.

What it really costs to run

The licence is free. Running the bot is not. Here is a realistic budget, with our own rough estimates where no official number exists:

Compared with a $29 to $129 monthly subscription on a hosted platform, Freqtrade is cheap in cash. It is expensive in hours. For a small account, the honest question is whether the time is worth it. Our trading bot ROI calculator lets you plug in hosting and fees to see the capital you need before any of this breaks even.

Security, API keys and custody

Self-hosting removes the biggest risk with commercial bots: a third party storing thousands of users' API keys in one database. That is exactly what happened in the 2022 3Commas leak. With Freqtrade, your keys sit in a config file on your machine. Nobody else holds them.

The flip side is that security is now your job. The basics:

• Create exchange keys with trading permission only, withdrawals disabled, and an IP whitelist set to your server.
• Use a sub-account holding only the bot's budget.
• Lock down the server: SSH keys, a firewall, and FreqUI either bound to localhost or behind a strong password and HTTPS.
• Restrict the Telegram bot to your own chat ID, because anyone who can command it can force trades.
• Never paste your config, with keys, into a forum post or an AI chat when asking for help.

Your funds stay on the exchange, so the exchange's own security and licensing still matter. Freqtrade does not hold custody of anything.

The overfitting trap (read this twice)

Machine learning and hyperopt share one danger: they are very good at finding patterns that existed only in your historical sample. A model with hundreds of features and a parameter search over thousands of combinations will almost always find a configuration that looks brilliant in a backtest. Live, it often falls apart.

Fewer features, longer out-of-sample tests, realistic fees and slippage, and the humility to delete a strategy that fails are worth more than any model choice. Our AI trading strategies guide covers which approaches survive this kind of testing.

Track record and community

Freqtrade has nine-plus years of public history, 55,681 stars and 11,534 forks on GitHub, and code pushed the day before our check. There is no company behind it to fold or pivot, which is more than many commercial bots can say after the 2026 shakeout. There is also no Trustpilot profile worth reading, because there is no customer relationship. Support happens through the documentation, GitHub issues and the community Discord, and it is better for people who ask precise questions with logs attached.

The one recurring complaint in community threads is not about bugs. It is that people expected the example strategies or FreqAI to be profitable by default. They are not, and the project has never said they were.

Pros and cons

Who should use Freqtrade, and who should skip it

Use it if you write Python, want to understand exactly why each trade happens, and see the learning curve as the point rather than the price. It is also the right tool if you want to experiment with ML on crypto data without paying a vendor that hides its methods.

Skip it if any of these apply: you want something running tonight, you have never used a command line, you expect the AI to find profits for you, or you will not monitor a server. In those cases a hosted platform from our beginner-friendly bots guide or the Bitsgap review is a better starting point. You lose real ML, but you also lose most of the ways to hurt yourself.

Freqtrade alternatives

For a fuller comparison, including licences and maintenance activity, see open-source AI crypto trading bots. If you want to understand where ML fits relative to rule-based grid and DCA logic, start with how AI trading bots work. Hosted reviews: Pionex, 3Commas.

Your bot needs an exchange. Pick one with checkable licences.

Freqtrade keeps your keys on your own server, but your funds still sit on an exchange. CEX.IO has operated since 2013 and publishes its registrations: a FinCEN MSB with money transmitter licences in 39 US states plus DC, and FCA registration in the UK. It is not on Freqtrade's official exchange list, so a sensible split is a regulated account for buying and long-term holding, and a separate, capped sub-account on a supported exchange for the bot.

Open a Verified Account Read our CEX.IO review

Registrations checked October 2026 on the company's legal pages. Availability depends on your country; crypto is high-risk.

• Since 2013One of the longest-running crypto exchanges still operating
  source
• 39 US states + DCFinCEN-registered MSB with state money transmitter licences (NMLS 1804170)
  source
• FCA-registeredUK cryptoasset business under the MLRs, FRN 1007192
  source
• PCI DSS Level 1Card-payment security standard, 2026 certificate published
  source

How to get started safely

Final verdict

Freqtrade with FreqAI earns its 4.4 because it is honest technology: real models, real retraining, open code, no fees and no promises. Among the dozens of products marketed as AI crypto bots, it is the clearest case where the label is earned. Its weaknesses are the mirror image of its strengths. Everything is in your hands, including every way to lose money.

If you can code and you treat backtests with suspicion, there is nothing better at any price. If you cannot, or you want results without the research, choose a hosted bot and come back to Freqtrade when curiosity wins. Either way, the rules of safe AI crypto trading apply: small sizes, trade-only keys, and no strategy you do not understand.

Frequently asked questions

Is Freqtrade free?

Yes. Freqtrade and its FreqAI module are free, open-source software under the GPL-3.0 licence. There is no subscription, token or paid signal tier. Your real costs are a server to run it on, exchange trading fees, and your own time to write, test and maintain strategies. Use our bot ROI calculator to see how fees and hosting affect a small account.

Is Freqtrade profitable?

Freqtrade itself is neither profitable nor unprofitable: it executes whatever strategy you give it. The project publishes no performance claims, which is a good sign. Most strategies that look great in a backtest degrade live because of overfitting, fees and slippage. Expect to spend weeks in dry-run before risking money, and read our evidence review Do AI crypto trading bots work?

What is FreqAI?

FreqAI is Freqtrade's machine-learning module. It trains regression or classification models such as LightGBM, CatBoost, PyTorch networks and reinforcement-learning agents on features you define, then retrains them periodically while the bot runs. It is described in a peer-reviewed paper in the Journal of Open Source Software (2022, DOI 10.21105/joss.04864). It is real ML, but it does not hand you a profitable model.

Is Freqtrade safe?

The code is open and widely reviewed, and your API keys stay on your own server rather than with a third-party platform. The risks sit with you: securing the server, creating exchange keys with trading permission only and withdrawals disabled, and protecting the Telegram bot and FreqUI login. A badly secured VPS is the most likely weak point.

Do I need to know Python to use Freqtrade?

Realistically, yes. Strategies are Python classes, configuration is JSON, and FreqAI feature engineering is code. You can start from the example strategies, and an AI model can help you write code, but you still need to read and debug it. If that sounds unappealing, a hosted bot from our beginner bots guide is a better fit.

Freqtrade vs Jesse vs Hummingbot: which should I use?

Freqtrade is the most complete all-rounder and the only one with a mature, documented ML module. Jesse has a cleaner research workflow but needs a paid lifetime licence for live trading. Hummingbot is built for market making and arbitrage with 318 connectors. Our open-source bots comparison covers each in detail.

Can Freqtrade trade futures?

Yes. As of October 2026 the docs list futures support for Binance, Bitget, Bybit, Gate, Hyperliquid, Kraken and OKX. Leverage magnifies both overfitting and liquidation risk, so test futures strategies in dry-run for longer than spot ones and keep leverage low. Our futures bots guide explains the liquidation maths.

Keep reading

AI bots Open-Source AI Crypto Trading Bots The open-source crypto trading bots on GitHub that are actually maintained, what each is good at, licences, and… Guide Build an AI Crypto Trading Bot Step by Step Build an AI crypto trading bot step by step: data with CCXT, features, a model, backtesting, paper trading and safe… Guide How AI Crypto Trading Bots Work Inside an AI crypto trading bot: data, signals, execution and risk layers, how ML bots differ from grid and DCA… Guide Do AI Crypto Trading Bots Actually Work? What evidence exists on AI crypto bot returns, why most performance claims cannot be verified, realistic… Calculator Trading Bot ROI Calculator Will a paid trading bot pay for itself? Enter capital, expected return, trading fees and subscription cost to see…

Sources and fact-check dates

We link to primary sources wherever possible. Dates show when the page was published or when we last checked it.

1. Freqtrade documentation (features, supported exchanges) · checked 2026-10-21
2. FreqAI documentation · checked 2026-10-21
3. FreqAI configuration docs (example config and command) · checked 2026-10-21
4. GitHub API: freqtrade/freqtrade repository stats · checked 2026-10-21
5. Caulk et al., "FreqAI: generalizing adaptive modeling for chaotic time-series market forecasts", JOSS · 2022
6. CFTC: AI Won't Turn Trading Bots into Money Machines · 2024-01-25