Step 1
Download
download Kwants library
Kwants
$700
$497 one-time · no subscription
Your LLM thinks it's backtesting. It's probably not.
Kwants plugs into any LLM — Claude, GPT, Gemini, Cursor — and gives it clean, point-in-time market data with lookahead bias blocked at the architecture level. No data engineering. No hallucinated results.
Status: idle
$ $
Powered By:
How it works
Five steps to your first backtest
Install, add your Polygon key, define your strategy in config, and run. No backtest engine to write.
Step 2
Config
Add your Polygon API key and download the data
Step 3
Describe
Tell your LLM what you want to test
Step 4
Run
kwants run — done
Step 5
Trade
Export results, move to live trading
why trust it
The reason your current backtest results are probably wrong
LLMs are great at describing strategies. They're terrible at enforcing data discipline. Left unconstrained, they'll peek at future data, ignore delistings, and hand you results that look great and perform terribly. Kwants blocks all of it at the data layer — not by convention, but by architecture.
Natural language, strict rules
Describe your strategy in plain English. Kwants enforces the constraints your LLM won't police itself.
Hallucination prevention baked in
Validation scripts catch logic errors LLMs generate with confidence. Your strategy runs what you described, not what the model assumed.
Point-in-time data, enforced
Data access is gated by the architecture. Signals can only see what was published before that day's open. No exceptions.
Polygon, without the hussle
Splits, adjustments, late prints, survivorship — all handled. Your LLM gets clean data, not Polygon's raw quirks.
Popular API: Polygon/Massive
LLM compatible
deterministic DSL
Assumptions, removed
Every LLM makes the same assumptions. We fixed them.
Survivorship bias, lookahead leakage, bad splits, late prints — most LLM runs ignore these. Every item below is live and running on every backtest you run.
what_we_fixed.sh
> ls -la ./fixes/
Lookahead bias
Live
survivorship bias - point in time accuracy
Live
no fake splits
Live
Late prints cleaned
Live
fast vectorized backtests
Live
no code but no hullecinations
Live
soon: backtest to live accuracy
Soon
soon: overfitting
Soon
soon: fundementals
Soon
Before you talk yourself out of it
Before you talk yourself out of it Common questions, straight answers
"Do I need to know Python?"
No. You describe your strategy in plain English to any LLM. Kwants handles the data layer. Python access is available for customization, but never required.
"$497 is a lot."
The real comparison isn't $497 vs. free. It's $497 vs. deploying a strategy with a lookahead bug in a live account. No recurring fees. Ever.
"What if my strategy isn't supported?"
Kwants is strategy-agnostic — momentum, mean reversion, intraday, daily. If you can describe it in plain language, you can test it. Reach out before purchasing if unsure.
FAQ
Quick Answers
A Python library that handles data downloading, processing, and backtesting — built to work with any LLM. You describe the strategy; Kwants handles the data integrity layer.
Any — ChatGPT, Claude, Gemini, Cursor, local models via Ollama. Kwants doesn't require a specific model.
No. One-time $497, no monthly fees, no seat licenses, no usage limits. You buy it once and own it permanently. You do need a massive.com subscription.
Data access is enforced at the infrastructure level. When a signal is generated for a given date, only data published before that date's market open is accessible. Gated by architecture, not convention.
No. Everything runs locally. No external requests are made by the library beyond massive.com API calls.
Stop Testing on Data That Lies.
Your strategy deserves a backtest that reflects what would have actually happened.
Get Kwants for $700$497- No code required
- Works with any LLM
- $497 one-time, own it permanently