Freqtrade vs Jesse

Choose Freqtrade for a fully open-source crypto workflow with free dry-run and live exchange execution, a broad maintainer-tested list of exchanges, and FreqAI's automated retraining. Choose Jesse for its route-based strategy structure, MIT research core, traditional-market candle research, research API, Monte Carlo and significance tests, or MCP workflow. Jesse's broader paper and live trading features require a separately licensed plugin.

Both are active, self-hosted Python tools with candle backtesting, Optuna-based parameter search, web interfaces, crypto spot and perpetual-futures support, and exchange execution. The main differences are licensing, live access, market scope, strategy organization, and research tools.

This comparison reflects Freqtrade 2026.8, released August 31, 2026, and Jesse 3.2.0, released September 17, 2026. Exchange support, prices, and plan limits were checked on September 19, 2026 and can change.

Freqtrade vs Jesse at a glance

License and total cost

Freqtrade is GPL-3.0 licensed, including its live-trading code. There is no separate product fee to enable a supported exchange. GPL is not a permissive license, so teams that distribute modified or combined software should review its rules. Hosting, market data, exchange fees, monitoring, and development still cost money.

Jesse uses the MIT License for its core research repository. Live and paper trading run through an official prebuilt plugin installed with an account license. The plugin is not part of the MIT grant.

As accessed on September 19, 2026, Jesse's pricing page listed lifetime Basic, Pro, and Enterprise plans at $899, $999, and $1,599. The free plan allowed one hourly or daily route on testnet exchanges but not paper trading. Paid plans enabled the documented live exchanges and paper mode, with different route, timeframe, IP, support, and research-feature limits. Verify both the current price and license terms before treating a lifetime price as the complete long-term cost.

The practical distinction is clear: Freqtrade keeps the source and supported production path open under GPL. Jesse makes the research core permissive but sells packaged live connectivity, entitlements, and support. Neither model is inherently superior. It depends on distribution plans, budget, desired support, and tolerance for maintaining exchange code.

Strategy model and multi-market organization

Freqtrade applies a strategy class to all pairs selected by a static or dynamic pairlist. The class defines vectorized indicator columns, entry and exit signals, and callbacks for sizing, prices, stops, leverage, and position adjustment. A strategy can trade many pairs. The old description of Freqtrade as "one strategy per pair" was incorrect. Running materially different strategies usually means separate bot processes and careful account separation, although producer-consumer and message interfaces can coordinate deployments.

Jesse routes map an exchange, symbol, timeframe, and strategy class. Data routes add non-traded series. Several routes can run together, subject to product-plan limits in live mode. This is useful when BTC uses one strategy on a five-minute timeframe while ETH uses another on an hourly timeframe. Routes do not eliminate shared-capital, correlation, or exchange constraints. Verify how simultaneous decisions, balance allocation, and position limits behave for the intended portfolio.

The better model is the one that matches deployment. A single policy applied to a changing universe fits Freqtrade naturally. A small, explicit matrix of strategy-symbol-timeframe assignments fits Jesse naturally. Avoid choosing on syntax alone. Prototype the actual number of markets, strategies, and accounts.

Market and exchange coverage

Freqtrade is a crypto bot. Its current maintainer-tested exchange table distinguishes spot, futures, margin mode, and on-exchange stop support for Binance, BingX, Bitget, Bybit, Gate.io, HTX, Hyperliquid, Kraken, Kraken Futures, KuCoin, OKX, and Bitvavo. CCXT exposes more exchanges, but the documentation explicitly separates that broad CCXT surface from exchanges tested by Freqtrade maintainers. Do not equate "CCXT supports it" with production support.

Jesse's current paid plan table lists production endpoints for Apex Omni, Hyperliquid, Bybit, Binance, Coinbase, Gate.io, Lighter, KuCoin, and Kraken across the documented spot and perpetual products. Its free plan is limited to testnet exchanges. Coverage should be compared by the exact venue and product needed, not by a raw exchange count. Account type, region, margin mode, hedge mode, stop orders, funding data, and historical candle availability can differ within one exchange brand.

Jesse 3.2.0 also documents candle research for traditional markets including stocks, currencies, indices, and individual futures contracts. This is backtesting support, not live trading through traditional brokers. Dividends, stock borrow, contract multipliers, continuous futures, margin schedules, and rollover require separate treatment. Freqtrade remains crypto-specific.

Backtesting and fill assumptions

Both engines replay candles rather than a historical order book. They can model fees and common order rules, but neither reconstructs queue priority, hidden liquidity, network delay, rejection paths, or the price impact of the simulated strategy.

Freqtrade's backtesting documentation defines entry, exit, stop, return-on-investment, trailing-stop, and same-candle priority assumptions. A normal entry signal fills at a later candle open, while custom prices can fill when they fall inside the candle range. Current exchange precision and limits can enter historical runs because historical rules are usually unavailable. Dynamic pairlists can also introduce current-universe bias.

Jesse's candle simulator supports market, limit, and stop orders, partial position changes, multiple timeframes, fees, and documented gap behavior. A price touched inside a candle is still not proof that a live resting order would fill. Traditional-market imports preserve closures and gaps, but the core accounting remains simplified for stock borrow and futures contracts.

Use the same standard for both products:

• state signal and fill timestamps,
• pin candle data, timezone, symbol mapping, and corporate-action treatment,
• include maker or taker fees, spread, slippage, funding, borrow, and impact where applicable,
• cap orders by realistic liquidity and exchange limits,
• test gaps, partial fills, stale data, rejected orders, and restarts, and
• reconcile dry or paper behavior with small controlled live orders before increasing risk.

If tick ordering, latency, or book depth determines the strategy, neither candle engine is the right primary simulator.

Optimization and validation

Freqtrade Hyperopt uses Optuna algorithms and currently starts with random combinations before using NSGAIIISampler to minimize a configured loss. Search spaces can cover signals, return-on-investment rules, stop loss, trailing stops, trades, and strategy-defined parameters. Hyperopt repeatedly runs the backtester, so its fill and cost assumptions flow directly into the winner.

Jesse's research optimizer uses Optuna with Ray for parallel trials. It accepts separate training and testing candles, optimizes the training objective, freezes the best parameters, and then calculates test metrics. Current objectives include Sharpe, Calmar, Sortino, and Omega. Jesse also provides candle and trade Monte Carlo tools plus rule-significance tests.

Apply the same warning to both. An internal test segment is no longer untouched after a researcher uses its result to change the strategy, objective, window, universe, or search space. Record every trial and redesign, inspect parameter neighborhoods, use chronological walk-forward evaluation, and retain a later prospective period for the final process. More optimization features increase research capacity and therefore increase the need for multiple-testing controls.

Machine-learning workflows

FreqAI is not merely a placeholder for calling a scikit-learn model. Its official feature list includes automated feature expansion, historical simulation of repeated retraining, background live retraining, model persistence and crash recovery, outlier handling, normalization, and producer-consumer fleets. Freqtrade is therefore the better fit when periodic model retraining is part of the live bot design.

Jesse now has a core machine-learning research workflow, so the old claim that it lacked an equivalent capability was stale. Its current documentation covers collecting labeled data from backtests, binary and multiclass classification, regression, stationarity, meta-labeling, and loading predictions in a strategy. It is closer to an explicit research pipeline than FreqAI's integrated adaptive retraining service.

Neither workflow makes a target or feature causal. Fit scalers, selectors, and models within each training interval. Align feature availability with decision time, preserve a label gap or purge where needed, and include failed model and stale-prediction behavior in dry or paper testing.

Dry run, paper trading, and live use

Freqtrade dry run is part of the open project. It consumes live market data, simulates orders, stores trades, and exercises much of the bot's operational path without sending exchange orders. Production mode uses the same general system with real credentials and orders. FreqUI, the REST API, Telegram, and webhooks support monitoring and control.

Jesse provides testnet live trading on its free plan. Paper mode and live exchange access require a paid plan. Its dashboard, notifications, routes, and live plugin provide a documented path from research to trading. Jesse also exposes an MCP server and rules for research clients. Any client that can run research or trading commands should use restricted credentials and independent risk controls.

Dry run, paper mode, and exchange testnets answer different questions. A local dry run checks bot timing and state. Paper mode checks more of the live data path. A testnet checks venue API behavior against an artificial market. None validates production liquidity, custody, funding, rate limits, or recovery from an exchange outage.

For either framework, use a dedicated subaccount, API keys without withdrawal permission, IP restrictions where supported, hard exposure and loss limits, clock synchronization, state reconciliation, database backups, alerts, and a tested shutdown procedure. Confirm that exchange positions and open orders can be reconstructed after a process, host, or network failure.

Installation and maintenance

Both projects treat Docker as a first-class path. Freqtrade's installation guide describes PyPI as an alternative that requires TA-Lib first and is not the recommended installation. Jesse also provides a first-class Docker project and can install its version-matched live plugin automatically when a license token is present.

Container use does not guarantee reproducibility. Pin image tags or digests, candle snapshots, Python dependencies, strategy source, configuration, exchange metadata assumptions, and database migrations. Test upgrades in backtest and dry or paper environments before changing production. Exchange APIs change independently of either release cycle.

As of the review date, both projects were actively releasing. Maintenance activity is necessary for exchange software but not enough to prove that a specific venue, account type, or operating system is trouble-free.

Which should you choose?

Choose Freqtrade when most of these are true:

• the system trades crypto candles on a documented Freqtrade venue,
• GPL-3.0 is acceptable,
• open-source dry and production execution are requirements,
• one strategy applied across a pairlist matches the deployment model,
• FreqAI's integrated historical and live retraining is valuable, and
• you prefer FreqUI, REST, Telegram, and its wider range of exchange settings.

Choose Jesse when most of these are true:

• explicit strategy-symbol-timeframe routes match the portfolio,
• the permissive MIT research core matters,
• traditional-market candle research is useful despite its accounting limits,
• built-in Monte Carlo, significance testing, research APIs, ML tutorials, or MCP workflows fit the process,
• the dashboard and supported commercial live path are worth the license cost, and
• the required venue, timeframe, route count, and IP count fit the chosen plan.

Choose neither as the sole engine when the edge depends on order-book state, queue priority, sub-candle event order, or latency. Also choose neither until the exact exchange, region, account type, market, and historical-data path have been verified.

A small test before you choose

Before committing, run the same small strategy and data fixture in both systems:

1. Use identical candles, fees, starting equity, position size, signal lag, and timestamps.
2. Force one market entry, limit fill, stop, gap, same-candle conflict, and rejected or unfilled order case.
3. Export every order, fill, fee, position, and equity transition and explain any difference.
4. Run the supported dry, paper, or testnet mode for the target venue and record restart behavior.
5. Measure setup time, iteration time, memory, logs, monitoring, backup, and upgrade effort on the intended host.
6. Review license and plan terms against distribution and production requirements.

That test is more informative than stars, generic feature counts, or a toy return. Freqtrade and Jesse can both support serious candle-based crypto systems. The better choice is the one whose licensing, market support, research process, and failure behavior match the system you will actually operate.