The production-grade Freqtrade fork for algorithmic trading on Hyperliquid.
Maintained by Freqtrade France.

Migrate from upstream · Start fresh · About · Features · Strategies · Learn · Full feature list

Freqtrade Ultimate is a drop-in superset of upstream freqtrade. Same freqtrade command, same configuration schema, same database format — it just adds features. Your bot is its user_data/ folder (strategies, configs, hyperopt and backtest results, downloaded data) plus its trade database. Switching forks means pointing a Freqtrade Ultimate install at those same files. Nothing about a running bot changes: same open trades, same config, same strategy — it simply resumes with extra capabilities available.

The safest path is non-destructive: install the fork in a new folder and run your existing bot from it. Your old install is never touched, so rollback is "just start the old one again".

# 1. Stop your current bot cleanly (Ctrl+C in its screen/tmux, or `systemctl stop`).

# 2. Back up your trade database (cheap insurance).
cp /path/to/tradesv3.sqlite /path/to/tradesv3.sqlite.bak

# 3. Clone Freqtrade Ultimate into a NEW folder — your old install stays as rollback.
git clone https://github.com/titouannwtt/freqtrade-ultimate.git
cd freqtrade-ultimate

# 4. Install it (creates its own .venv; your old environment is untouched).
./setup.sh -i

# 5. Activate and confirm you're on the fork.
source .venv/bin/activate
freqtrade --version

# 6. Run YOUR bot with YOUR existing files — nothing copied, nothing lost.
freqtrade trade \
  --config   /path/to/your/config.json \
  --strategy YourStrategy \
  --userdir  /path/to/your/user_data \
  --db-url   sqlite:////absolute/path/to/your/tradesv3.sqlite

Prefer one tidy folder? Instead of pointing at old paths, copy your data into the clone: cp -a /path/to/your/user_data/. ./user_data/ then place your config.json and *.sqlite where you normally keep them, and launch as usual.

There is no public Ultimate image — you build it yourself from source (the repo's .dockerignore excludes the bundled data, so the build stays small and fast). Recommended path: keep your working docker-compose.yml, volumes and DB exactly as they are, and swap only the image.

# 1. Stop your current stack.
docker compose down

# 2. Build the Freqtrade Ultimate image from source.
git clone https://github.com/titouannwtt/freqtrade-ultimate.git
cd freqtrade-ultimate
docker build -t freqtrade-ultimate:latest .
cd -

# 3. In your EXISTING docker-compose.yml, change the image line:
#       image: freqtradeorg/freqtrade:stable
#    to:
#       image: freqtrade-ultimate:latest

# 4. Bring the bot back up — same mounted user_data, same DB, same trades.
docker compose up -d

Your user_data/ volume and database are never rewritten by this — only the code inside the container changes. Rollback: set the image line back to freqtradeorg/freqtrade:stable and docker compose up -d. Update the fork later: git pull in the clone → docker build -t freqtrade-ultimate:latest . → docker compose up -d.

You don't need to install upstream freqtrade first — Freqtrade Ultimate is a complete, ready-to-run installation on its own.

🇫🇷 Francophone ? La communauté Freqtrade France propose un guide écrit pas-à-pas (installation, configuration, premier bot) : Freqtrade — le guide complet du bot de trading open source. (French only / en français uniquement.)

git clone https://github.com/titouannwtt/freqtrade-ultimate.git
cd freqtrade-ultimate
./setup.sh -i

source .venv/bin/activate
freqtrade create-userdir --userdir user_data
freqtrade new-config --config user_data/config.json

The bundled docker-compose.yml builds the Ultimate image for you (tagged freqtrade-ultimate:latest):

git clone https://github.com/titouannwtt/freqtrade-ultimate.git
cd freqtrade-ultimate

# Build the image from source.
docker compose build

# Create your user_data folder and a starter config:
docker compose run --rm freqtrade create-userdir --userdir user_data
docker compose run --rm freqtrade new-config --config user_data/config.json

# Edit user_data/config.json (exchange, stake, pairs), then launch:
docker compose up -d

Next steps: edit user_data/config.json (exchange, stake, pairs), drop a strategy into user_data/strategies/, then launch with freqtrade trade (native) or docker compose up -d. The full Freqtrade documentation applies unchanged; see the Learn section below and docs/FEATURES.md for what this fork adds.

A maintained, opinionated fork of Freqtrade optimized for serious algorithmic trading on Hyperliquid perpetual futures, with 35+ features not present upstream.

Why this fork exists. Running multiple Freqtrade bots in production on Hyperliquid surfaces real-world problems upstream wasn't designed for — rate-limit cascades when four bots refresh OHLCV simultaneously, ADL and liquidation handling on a DEX without traditional liquidation events, multi-bot pairlist deduplication, and statistically valid hyperopt without curve-fitting. This fork solves those.

Editorial principle: no curve-fitted strategies. Every showcase strategy in this repo ships with its walk-forward analysis and real drawdowns. We do not promote backtest-pretty strategies that fail live — many popular Freqtrade strategies embed subtle lookahead biases that make backtests look magical and live results disappointing. We document why, and we publish the methodology that avoids it.

A quick tour. Full inventory with implementation details lives in docs/FEATURES.md.

• OHLCV Cache Daemon (ftcache) — Shared candle cache across N bots, 75 % API call reduction measured in production with four bots.
• Pairlist Cache Daemon (ftpairlists) — Deduplicates pairlist filter computation between bots (pairlist refresh: 15 min → 3 min).
• Position Guard + Leverage Sync — Prevents conflicting entries and unintended hedges on shared wallets.
• Fleet State Notifications + Auto-Restart — Thundering-herd prevention with startup jitter, launch_bot.sh auto-restart loop.

• PlateauSampler — Coordinate-wise Optuna sampler for robust hyperparameter optimization (four-phase: baseline → scan → assembly → refinement).
• --sampler CLI flag — Switch between TPE, NSGA-II/III, CMA-ES, GP, QMC samplers without editing your strategy code.
• Walk-Forward Analysis (freqtrade walk-forward) — Rolling, anchored, and CPCV (Combinatorial Purged Cross-Validation) modes, plus Monte Carlo drawdown simulation, PBO (Probability of Backtest Overfitting) score, verdict A–F, and an interactive HTML report.
• Custom hyperopt losses — MoutonMeanRev (mean-reversion / DCA), MoutonMomentum (trend / momentum), MyProfitDrawdown (simple baseline).

• Liquidation detection via user-fills monitoring (liquidationMarkPx).
• External close detection (ADL or manual UI close) with exit_reason="external_close".
• Resilient leverage / margin error handling in DCA continues operation instead of crashing.
• Local Hyperliquid historical data bundle — 3 250+ Feather files (300+ pairs × 8 timeframes including crypto indices and TradFi perpetuals).

• TrendRegularityFilter — Excludes pairs with a regular linear uptrend (essential for short strategies).
• backtest_lock_wallet — Disables compounding in backtests for honest equity curves.
• Capital withdrawal accounting — Tracks net profit after capital removal in REST API, Telegram and /profit.

• ExchangeMetrics — Ring-buffered API-call metrics, 429 tracking, live token-bucket state.
• REST API enriched — /cache_status, /rate_metrics, /fleet/status, /fleet/events, /volume_history, /signal_summary, /stratdev/* (consumed by frequi-ultimate).
• Enhanced Telegram — LIQUIDATION and external_close exit reasons; withdrawal-aware /profit.

• Replay engine — Drives FreqtradeBot.process() (the real live code path) candle-by-candle over historical data via a virtual clock and fake exchange overlay. Validate a strategy in hours instead of months.
• 1-minute resolution, real funding — Stoploss/ROI/signals checked every virtual minute; funding computed from local Feather files. Equivalent to permanent --timeframe-detail 1m on the real engine.
• Config auto-launch — Add a dry_run_replay block to your bot config and the bot auto-seeds a 5-month replay on first start, then transitions to normal dry-run. Zero manual action.
• Coordinator daemon — Caps concurrent replays to nproc - 2 - hyperopt_cores with SIGSTOP/SIGCONT priority queue. Auto-spawned.

• Strategy Dev Backend — Reader, jobs runner, strategy editor APIs (consumed by frequi-ultimate).
• AI Copilot — Repository ships with CLAUDE.md plus 14 tip files containing 199 trading rules curated from Carver, Clenow, Chan, López de Prado, and the Freqtrade France community.
• Enhanced CLI help — Practical guidance, recommendations, and tradeoffs documented in every option.

➡️ Full feature list with deep technical details: docs/FEATURES.md

This repository ships with public showcase strategies directly inside user_data/strategies/. Each strategy follows a strict naming convention:

These strategies are intentionally simple and honest about their limits. They demonstrate the methodology (anti-overfitting, walk-forward, real drawdowns), not maximum profitability. More advanced and live-tested strategies are reserved for Freqtrade France members along with full live PnL screenshots and reproducible parameters.

Installation is covered above — migrate from upstream or start fresh. This fork is a drop-in replacement: all upstream commands work unchanged, and the full Freqtrade documentation applies. Fork-specific commands and flags are documented in docs/FEATURES.md.

For the multi-bot UI optimized for this fork (50+ enhanced components, fleet comparison, market context):

freqtrade install-ui --ui-version github://titouannwtt/frequi-ultimate

Or visit titouannwtt/frequi-ultimate.

Freqtrade France is the French-speaking community where Mouton (this fork's maintainer) publishes:

• 📚 Free tutorials — Freqtrade basics, Hyperliquid setup, hyperopt, backtesting, walk-forward analysis (80 % of the content is free).
• 💎 Member tutorials (9 € / month or 90 € / year) — PlateauSampler internals, custom hyperopt loss design, walk-forward CPCV deep dives, anti-overfitting playbook.
• 🤖 Ready-to-deploy strategies for members — Live-tested, with reproducible backtests and live PnL.
• 🎥 Long-form YouTube — @freqtrade_france.
• 🐦 Twitter — @MoutonCrypto.

If you don't want to subscribe but want to support the fork, the simplest free way is to use the Hyperliquid referral link when creating your account.

PRs are welcome on infrastructure features (caching, hyperopt samplers, walk-forward, observability). See CONTRIBUTING.md. Strategy-specific contributions and discussions happen in the Freqtrade France community.

This is educational software. Past performance does not guarantee future results. You are responsible for your own trading and any losses incurred. This project does not provide investment advice. Trading crypto futures is high risk and can lead to total loss of capital.

GPL-3.0 — same as upstream Freqtrade.

Built and maintained by Mouton 🐑 · Freqtrade France