QuantConnect, Freqtrade, Hummingbot or an AI agent: which for crypto quant trading?

Which should a crypto quant use: QuantConnect, a Python framework or an AI agent? QuantConnect is the deepest research and backtesting platform, but it does not support Hyperliquid: its crypto perps run on Binance, Bybit and dYdX. On Hyperliquid itself, pick a framework if you want to write and run the code, or an AI agent if you want to describe the strategy and have it run for you. Among the frameworks, for one codebase that backtests and trades live, NautilusTrader: it has a native Hyperliquid adapter for spot and perps, with HIP-3 markets as an opt-in. For a ready-made strategy bot with backtesting and parameter optimisation, Freqtrade, which reaches Hyperliquid through CCXT. For market making, Hummingbot, which has native spot and perp connectors and runs paper trades on Hyperliquid’s testnet. For your own code with one API across many venues, CCXT. backtrader and vectorbt are research and backtesting tools with no Hyperliquid connection of their own.

QuantConnect, Python frameworks and an AI agent compared

QuantConnect

QuantConnect is a hosted platform built on LEAN, its open-source engine (Apache-2.0). You write the algorithm in Python or C#, with its AI assistant Mia if you like, backtest it on QuantConnect’s data with fee and slippage models, and rent a live node that trades through one of 20-plus brokerages. Research is free; live trading starts with the Researcher plan at $84 a month. Hyperliquid is not among its brokerages or data sources, so it suits strategies you trade elsewhere. The full comparison is in QuantConnect vs Senpi.

NautilusTrader

NautilusTrader is an event-driven engine with a Rust core and a Python API, so the strategy you backtest is the one you run live. Its Hyperliquid adapter covers spot and perps, with HIP-3 and HIP-4 markets as an opt-in in the current stable release, v1.231.0 (Aug 2026). The v2 line, still in release candidates on Nov 13, 2026, loads spot, perps and HIP-3 markets at connect and labels them stable; its README advises against using release candidates for live trading with real money. Mainnet orders carry NautilusTrader’s builder code at a zero fee rate, after a one-time approval, and you can turn the attribution off. Its issue tracker is busy with Hyperliquid work: about a dozen issues opened between Oct 16 and Nov 10, 2026 concern the adapter, mostly reconciliation after a restart, account equity and pre-trade margin checks. Test your exact order types on testnet first.

Freqtrade

Freqtrade is a complete strategy bot: you write a strategy class, backtest it, optimise its parameters and run it live, with a web or Telegram front end. It reaches Hyperliquid through CCXT and supports futures, spot, sub-accounts and vaults, and it places stop-losses on the exchange. Two Hyperliquid details matter. Hyperliquid has no native market order, so CCXT simulates one with a limit order at up to 5% slippage. And the API serves only 5,000 historical candles, so a backtest covers 5,000 bars unless you download candles over time. Freqtrade’s docs recommend an API wallet key, which can trade but not withdraw, and a sub-account the bot owns alone.

Hummingbot

Hummingbot is built around market making and arbitrage. It has native Hyperliquid spot and perp connectors, HIP-3 markets on the perp connector (isolated margin only), backtests for its V2 controller configs, and paper trading through Hyperliquid’s testnet on perps; its own client tags orders with a 0 bps builder code. If you want to quote both sides of a book, start here; How market making works on Hyperliquid covers the economics.

CCXT

CCXT is a library, not a strategy engine. Its hyperliquid class gives you the same calls you use on other venues (markets, order books, candles, orders), with spot, perps and HIP-3 markets and a sandbox mode that points at testnet. Use it when your strategy logic is your own and you want one API across Hyperliquid and other exchanges; Freqtrade itself uses it underneath.

backtrader and vectorbt

backtrader is a long-standing event-driven backtester with a broker simulation, but it has no Hyperliquid connection and no code change since April 2023, so live trading on Hyperliquid means writing your own broker or data feed. vectorbt is a vectorised research library for testing many parameter sets quickly. It does not trade live, and its licence is Apache 2.0 with a Commons Clause; its PRO edition is paid. Both are good for research on Hyperliquid data you load yourself, and How to backtest a crypto trading strategy covers costs and pitfalls.

What every framework has to handle on Hyperliquid

• Rate limits. REST requests share 1,200 weight a minute per IP. Order actions are also limited per address: 1 request per 1 USDC traded since the address was created, plus a starting buffer of 10,000.
• Prices and sizes. Prices allow 5 significant figures and at most 6 decimals minus the market’s szDecimals on perps; sizes round to szDecimals.
• Websockets. The server closes a connection with no message for 60 seconds, so send {"method": "ping"}. One IP can hold 10 connections and 1,000 subscriptions.
• Keys. Bots sign with an API wallet, which can trade but not withdraw, on behalf of your main address.
• HIP-3 markets. Builder-deployed markets carry a prefix, such as xyz:NVDA, and some calls need the market’s dex name. See What is HIP-3?

Hyperliquid’s official Python SDK has a worked example for each of these calls, and Python algorithmic trading for crypto walks through a first bot.

Framework or AI agent?

Where Senpi fits

A framework gives you full control and leaves you to run the server, the keys, restarts and monitoring. Senpi is the other route: you describe the rule in plain language, and Senpi’s AI quant writes it as a strategy and runs it. On Senpi, a strategy you confirm runs as code on Senpi’s runtime, and by default no AI model makes live trading decisions. Strategies built from a rule or a template carry the Dynamic Stop-Loss (DSL) from the first fill, and each can have its own daily-loss halt, which lifts at the next UTC midnight, or a drawdown halt, which waits until you resume it. Copy strategies carry at most a flat stop. Each strategy trades from its own strategy wallet; the agent holds no key and cannot withdraw.

Senpi doesn’t run backtests; Simulations, paper trading with simulated money, are coming. Today a strategy is tested with a validation run and then a small live budget. Senpi’s strategy templates are open source in the senpi-skills repository on GitHub, so you can read exactly what each one does. Claude Code and Codex can add Senpi’s agent as a custom connector, and Claude and ChatGPT reach it through Senpi Connectors. Hyperliquid charges 0.015% for maker orders and 0.045% for taker orders at the base tier on crypto perps. Senpi adds 0% on paid plans and 0.03% on the free trial or without a paid plan.

This is educational material, not financial advice. The market readings are data, not trade signals, and leveraged perps can lose money quickly.