Open-Source AI Crypto Trading Bots

Most commercial bots sell you "AI" you cannot inspect. Open-source bots are the opposite: every line is on GitHub, the backtests run on your machine and the API keys never leave your server. The price is effort. You need some Python, a server and the patience to test properly.

This page covers the open-source projects that are actively maintained, what each is good at, how they compare, what it costs to run them and how to avoid the fake "AI crypto trading bot" GitHub repos that exist only to steal from you. If you would rather use a hosted product, start with our ranking of the best AI crypto trading bots.

Why choose an open-source AI crypto trading bot?

Three reasons, in order of importance. Custody of keys: a hosted platform stores your exchange API keys on its servers. The 2022 3Commas leak, in which roughly 100,000 keys were reportedly exposed, showed what can happen. A self-hosted bot keeps keys on a machine you control. Transparency: you can read exactly how a signal is computed and how orders are placed, so there is no "proprietary AI" black box. Cost: the software is free. You pay for a server and exchange fees, not a monthly subscription.

The downsides are real too. Nobody will fix a broken strategy for you, there is no support desk, and a misconfigured bot can trade badly 24 hours a day. Open source protects you from the vendor, not from yourself.

Freqtrade + FreqAI: the real machine-learning option

Freqtrade is a Python bot started in 2017 and licensed under GPL-3.0. As of 21 October 2026 its repository had about 55,700 stars and 11,500 forks, with commits the day before we checked. It supports spot trading on Binance, BingX, Bitget, Bybit, Gate, HTX, Hyperliquid, Kraken, OKX and MyOKX, and futures on several of those, plus other venues through the CCXT library without guarantees.

The core bot gives you backtesting, hyperparameter optimisation (hyperopt), dry-run paper trading, a web UI called FreqUI and Telegram control. Strategies are Python classes you write or adapt.

FreqAI is what makes it an AI bot in the real sense. It automates training of regression and classification models (LightGBM, CatBoost, PyTorch, reinforcement learning, with CNN examples) on features you define, and it retrains them periodically during live and dry-run trading so the model adapts as markets change. Backtests simulate that retraining too. The approach is described in a peer-reviewed paper by Robert Caulk and co-authors, "FreqAI: generalizing adaptive modeling for chaotic time-series market forecasts", published in the Journal of Open Source Software in 2022 (DOI 10.21105/joss.04864). The project is not-for-profit: no token and no paid signals.

Jesse: research-first, paid for live trading

Jesse is an MIT-licensed Python framework, created in 2018, with about 8,700 GitHub stars. It is known for clean strategy syntax and accurate backtests, with optimisation, Monte Carlo analysis, a research module, machine-learning tooling and MCP agent workflows (100 per day free) for driving it from AI assistants.

The important detail many comparison posts miss: the free tier is for backtesting and testnet only. Live trading on real exchanges requires a paid lifetime licence: Basic $899 (1 IP, 1 trading route, hourly and daily timeframes), Pro $999 (2 IPs, 10 routes, all timeframes) or Enterprise $1,599 (5 IPs, 50 routes). Supported live venues include Binance, Bybit, Coinbase, Kraken Pro, KuCoin, Gate.io, Hyperliquid, Apex Omni and Lighter. The code is open; live execution is a commercial product.

Hummingbot: market making and arbitrage

Hummingbot is Apache-2.0 licensed, governed by the community-owned Hummingbot Foundation, and had about 20,600 stars when we checked. Its strength is breadth: 318 connectors across centralised exchanges and DEXs. It is designed for market making, where the bot quotes both sides of the order book and earns the spread, and for arbitrage between venues. See our arbitrage bots guide for why that is harder than it sounds.

The AI angle is Condor, described by the project as a harness for building and running agentic strategies and bot instances, alongside a Hummingbot API server. The project reports $47B+ in total volume and 111K+ instances on its public dashboard; those are self-reported figures.

ElizaOS and other agent frameworks

ElizaOS is an MIT-licensed "agentic operating system" for building AI agents, with about 19,600 stars and active commits in October 2026. It is popular for on-chain agents that hold wallets and act on their own, but it is a framework, not a trading strategy. You would still need to write the trading logic and the risk limits. Its old website now redirects to Eliza Research.

You will also see OpenClaw mentioned in trading circles. It is an MIT-licensed, self-hosted general personal assistant, not a crypto product, and users bolt trading "skills" onto it themselves. We cover the risks in OpenClaw for crypto trading, and agent tools more broadly in AI crypto trading agents.

We have left out other projects whose licence, maintenance and features we could not verify for this update.

Freqtrade vs Jesse vs Hummingbot

Our short answer: if you want an AI bot that actually learns, pick Freqtrade. If you mainly want to research ideas with clean code and are willing to pay for live execution, pick Jesse. If you want to provide liquidity or trade across many venues including DEXs, pick Hummingbot. Plenty of developers prototype in one and run in another.

Hosting and running costs

The software is free; running it is not quite. A typical setup is a small Linux virtual server running Docker, close to your exchange's servers if latency matters. A plain rule-based Freqtrade or Jesse bot runs comfortably on modest hardware. FreqAI is heavier: model training needs more CPU and RAM, and deep-learning models benefit from a GPU. Hummingbot market making is sensitive to latency and uptime.

Compare that with hosted subscriptions of $20 to $150 a month. For small accounts, the server is often cheaper than a subscription; the bigger cost is the hours you spend. The trading bot ROI calculator shows how fees and fixed costs eat into returns.

Security: your keys, your server, your responsibility

Self-hosting removes the third-party key store, but it moves the job to you. A sensible baseline:

• Create API keys with trade permission only and withdrawals disabled, and whitelist your server's IP address on the exchange.
• Use a dedicated sub-account that holds only the bot's budget.
• Do not expose the bot's web UI or REST API to the open internet. Use a VPN or SSH tunnel, a strong password and a firewall.
• Keep secrets in environment variables or a config file outside version control. Never commit a config with keys to GitHub, even a private repo.
• Update regularly, read release notes, and pin versions so an update never surprises a running bot.
• Set alerts (Telegram works well) for errors, unusual trade counts and drawdown limits.

AI crypto trading bot GitHub red flags

Search GitHub for an AI crypto trading bot and you will find thousands of repos. A few are excellent. Many are abandoned student projects. Some are traps. In February 2025 Kaspersky reported a campaign it called GitVenom: hundreds of fake GitHub projects posing as automation tools, crypto bots and utilities, with AI-generated README files and artificially inflated commit counts. The code pulled in an info-stealer, remote-access tools and a clipboard hijacker that swapped copied wallet addresses for the attackers' own. One attacker wallet reportedly received about 5 BTC.

Good signs are the opposite: a long commit history from many contributors, an open issue tracker with real discussion, a clear licence, documentation that explains risk, and no promise of returns. When in doubt, run unfamiliar code in a throwaway virtual machine with no wallets on it. For broader scam patterns, read our AI crypto trading scams guide.

Ready to build? Our step-by-step guide to building an AI crypto trading bot shows how to combine a model, deterministic execution and guardrails. Nothing here is financial advice; automated trading can lose money quickly.