Describe a strategy in plain words. AI builds it as a safe, structured spec. The backtest proves it on real market data. Only strategies that survive make it through — natural selection, automated.
A complete quant workstation you run on your own computer — no cloud, no lock-in, no monthly fee.
Pulls real OHLCV from 6 crypto exchanges (ccxt) and forex pairs (yfinance) — 8 major coins, any timeframe. Simulates trades with realistic fees + slippage and scores return, max drawdown, Sharpe and win rate against a Buy & Hold line, so you see whether the strategy actually beat just holding. Look-ahead bias guarded by design.
Describe a strategy — “buy when RSI drops below 30, sell above 70.” Claude turns it into a structured JSON spec (RSI, MACD, Bollinger, crossovers, multi-timeframe + multi-asset confluence) — no raw code runs, so it's safe to share. Tune parameters with grid-search optimization; save, refine and export/import strategies as JSON. Your API key stays in your browser.
An agent that edits the bot's own code, backs up before every change, health-checks after restart, and auto-rolls-back if anything breaks. Every change is auto-committed to git with a visible diff. It never edits itself — so it can't take itself down.
You don't write code. You set the direction; DARWIN mutates, tests, and keeps only what performs.
Tell it what to try, in plain language or a tweak to an existing strategy.
AI rewrites the strategy spec (and, in builder mode, the code) — backed up first.
The engine runs it on real history and scores it: return, drawdown, Sharpe, win rate.
Keep the survivors in your library; auto-rollback the ones that break. Repeat.
No mockups — this is the real tool you run on your own machine. Click any shot to enlarge.

Real OHLCV from 6 exchanges + forex. Realistic fees & slippage, full risk controls (position size, risk-per-trade, ATR / swing stops, trailing). Every result is measured against a Buy & Hold line — so you see the honest truth, not a cherry-picked number.

Grid-search tunes parameters and risk, then validates the winner out-of-sample. The red OOS column exposes overfitting at a glance. The market scan hunts statistical edges (RSI bounces, seasonality, volatility) and labels each "holds ✓" or "fails ✗" on unseen data.

Describe a strategy in plain words; Claude returns a structured JSON spec (indicators, entry/exit, risk) — no raw code ever runs, so it's safe to share. The self-evolving loop iterates and honestly reports when a winner does NOT hold out-of-sample.
Your Anthropic key and any exchange keys live on your machine. We never see, log, or store them.
The whole thing is a Python app you launch yourself. No account, no cloud dependency, no data leaves your computer.
AI returns structured specs, not arbitrary code. The builder backs up, health-checks, and auto-rolls-back. Hard to break, easy to recover.
Every strategy is a readable JSON spec you can export, share, edit, or run elsewhere. You own everything you build.
“Time in the market beats timing the market.”
— Warren Buffett. Most strategies lose to simply holding once you count fees and slippage. DARWIN puts the Buy & Hold line on every backtest, right next to your strategy — and tells you which one won. Don't take our word for it. Backtest it yourself.
DARWIN is a research and backtesting tool. It is NOT financial advice, NOT a signal service, and NOT a promise of profit. Nothing here is a recommendation to buy or sell any asset.
Backtest results do not guarantee future performance. Markets change, strategies overfit, and a curve that looked great on history can lose money live. Crypto trading carries substantial risk — you can lose your entire capital.
Only trade with money you can afford to lose, and never deploy a strategy live without thorough independent testing. You are solely responsible for your own trading decisions.
DARWIN is a hobby project still in the works. Leave your email and we'll reach out when there's something to try — no spam, no promises.
We'll email you when there's something to try.