Quantitative Research Laboratory

You have the intuition.
We built the lab to prove it.

TheoryCraft is an AI research platform for traders and investors. Describe an idea in plain words, and it builds the strategy and backtests it against years of real market data, so you can see whether your edge is real before you risk money.

Get your Access
1,600+ instruments 15+ yrs tick data No code required
theorycraft · validation
XAU/USD · London breakout Validated
1.51Profit factor
1.18Sharpe
-7.1%Max DD
47.2%Win rate
2010 to 2025 · spread, slippage and commission modeled overfit 18/100
theorycraft · exploration · PEA SP500 DCA Live
A TheoryCraft research notebook with strategy research charts

A live exploration notebook in TheoryCraft, comparing DCA vs lump sum strategies on the S&P 500.

The problem

You think you have an edge.
But do you really?

Most traders rely on gut feel and a handful of screenshots. Without enough data behind a strategy, you are trading on hope instead of evidence.

01 ✕
No rigorous way to validate. Testing manually on charts can't separate a real edge from random noise.
02 ✕
Not enough data to trust results. 50 journal trades or 6 months of backtest can't support a conclusion.
03 ✕
Testing takes weeks, not minutes. Building the infrastructure from scratch burns your iteration loop.
04 ✕
Great on paper, fails live. Overfitting stays invisible without proper out-of-sample testing.
05 ✕
Skill, or luck? Without statistical proof you'll never know which one you're holding.
The scientific method, applied to markets

Observation becomes proof

Every profitable system starts from a repeatable market event called an Edge. TheoryCraft helps you isolate that Edge and measure how reliable it is, then build filters and risk rules on top of it. Each layer is validated on its own before you add the next.

Edge

Your idea, tested

Find a repeatable pattern and measure its real probability of success.

+
Filters

Better entries

Keep only the highest-quality setups. Each filter is tested to prove it actually helps.

+
Risk

Capital protected

Sizing, stops and drawdown caps, verified to improve the strategy.

=
Strategy

Ready to deploy

An engineered, tradable framework backed by data you can trust.

Institutional-grade validation

Every strategy comes back with a verdict

The same methods hedge funds use to confirm an edge is real: walk-forward analysis, Monte Carlo simulation, and overfitting checks. This is the screen you will live in.

Walk-forward analysis

Tested out-of-sample on rolling windows. Does the edge hold on data it has never seen?

Monte Carlo

Thousands of resampled paths show the full range of possible outcomes, so you do not bet everything on one lucky run.

Overfit score

18 / 100 · low curve-fit risk

One honest number: how much of the performance is real, and how much is curve-fitting?

From idea to proof: four steps

Four steps to a strategy you can trust

01

Describe your idea

Say it in plain words, or write the logic yourself. Either way, it becomes testable.

02

Refine & protect

Add filters and risk rules. Each addition is tested so you keep only what helps.

03

Optimize & validate

Find robust parameters across years and assets, then run out-of-sample validation.

04

Incubate & deploy

Watch it run live before risking real money. Deploy only when the numbers earn it.

The research notebook

Readable code you own

You do not need to write code, yet every strategy is real, readable Python that you fully own. What the assistant writes is exactly what a developer would write, so you can audit it and run it on your own infrastructure.

  • Open, reproducible notebooks
  • Live charts & data exploration
  • Export & self-host anytime
gold_breakout_research.ipynb · connected
In [1]
# Hypothesis: London-session breakouts on XAU/USD have an edge
strat = Edge.breakout(asset="XAU/USD", session="london", lookback=20)
In [2]
res = backtest(strat, data="2010..2025", costs=True)
res.summary()
Out [2]
trades: 1,284 win%: 47.2 profit_factor: 1.51 sharpe: 1.18 max_dd: -7.1% edge: CONFIRMED ✓
Plot
research chat · TheoryCraft AI
You
Does adding an ATR volatility filter improve the gold breakout?
AI
Testing two thresholds across 2010-2025. Filtering setups below ATR<0.4% raised profit factor 1.38 -> 1.51 and cut max drawdown to -7.1%. Want me to walk-forward validate it?
You
Yes, and check for overfitting.
AI
Holds across 6 of 7 windows. Overfit score 18/100. Robust. I've saved the notebook to your workspace.
AI research assistant

Your research partner
in the lab

The assistant creates notebooks, writes the code, runs your backtests, and reads the results with you. It does not place trades or send buy and sell signals. Its job is to turn your question into evidence you can check yourself.

  • Plain-language idea → working, testable strategy
  • Reads results and suggests the next iteration
  • Bring your own agent through MCP, including Claude and Codex
Open by design

No black boxes.
The engine is yours.

The core engine is fully open source (Apache 2.0) and free forever. Every algorithm that touches your data is transparent and auditable. The platform adds AI, optimization and analytics on top.

Star on GitHub View the engine

Core engine

Tick-level backtest engine. Open, verifiable, built for accuracy.

Connectors

Open integrations for brokers, exchanges & data feeds.

Built for trust

Audit, extend or fork any module. The AI uses the same libraries you do.

Build your edge alongside other researchers

Discord is where the project lives. You will find daily research discussion, build updates, and a direct line to the founder.

Join the Discord
ImNotAVirus
Founder story

I thought profitable trading was about finding the perfect strategy. I was wrong.

I started trading in 2020, and I went through the same loop most beginners know too well : gambling on setups, jumping from one YouTube strategy to the next, listening to the wrong people, and paying gurus who probably were not profitable either.

I wasted 3 years thinking I was one strategy away. Every new setup felt like the answer for a few days, then one bad week made me doubt everything again, and I was back to searching for the next thing.

The painful part was not losing trades. It was having no honest way to know whether an idea was bad, badly tested, or just missing enough data. So I kept confusing hope with evidence.

That is the problem TheoryCraft solves: you take a trading idea, turn it into a real test on market data, and see if it deserves more of your time. A weak idea should be killed before it costs you more time and money .

ImNotAVirus
Founder and Trader, builds TheoryCraft
Pricing

Join early. Keep founder pricing.

One plan at launch, with full access to the platform. Early members lock in founder pricing while the product is still taking shape, and they keep that price as it grows.

Managed: AI included

Managed

$48.93/month $69.90 $489.30/year $699
Get Managed Get Managed

For researchers who want the full platform plus managed AI access from day one. No separate LLM account or API key required.

What's included
  • Full platform & all Lab features
  • Integrated research notebooks
  • AI research assistant with LLM access included
  • Platform-managed AI access
  • Historical data, 1,600+ instruments
  • Multi-year backtesting
  • Public notebook sharing
  • Ready-to-use research workflow
Founding bonuses
  • Founder price fixed for life
  • A direct line to the founder via Discord
  • OG Discord role to show you were here early
  • Early access to every new feature
BYOK: bring your own key

BYOK

$14.95/month $29.90 $149.50/year $299
Get BYOK Get BYOK

For researchers who already have AI provider accounts and want to plug their own keys into TheoryCraft.

What's included
  • Full platform & all Lab features
  • AI research assistant with your own LLM key
  • 26 AI providers: OpenAI, Anthropic, Gemini, Groq, Mistral...
  • Keep full control of model choice and API costs
  • LLM access is not included in BYOK
Founding bonuses
  • Founder price fixed for life
  • A direct line to the founder via Discord
  • OG Discord role to show you were here early
  • Early access to every new feature

Managed includes AI access. BYOK is for researchers who want to use their own provider keys.

FAQ

Questions,
answered

Still curious? Ask us anything on Discord.

What is TheoryCraft?
TheoryCraft is an AI research platform for traders, investors, and builders who want to stop relying on gut feeling, screenshots, or small sample sizes. It helps you turn a market idea into a testable strategy, run it on historical data, and decide whether the edge is worth taking seriously.
Is TheoryCraft an AI trading bot?
No. TheoryCraft does not place trades or generate automatic buy and sell signals. It is built for research, backtesting, validation, and strategy improvement before risking real money.
Do I need to know how to code?
No. You can describe an idea in plain English, and the AI assistant can help create the notebook, write the code, run the backtest, and explain the results.
How is TheoryCraft different from a regular backtesting platform?
Most backtesting tools give you a form or a coding environment. TheoryCraft gives you a full research workflow: an AI assistant, readable notebooks, historical data, validation methods, and the ability to iterate from a simple idea to evidence you can inspect.
What are walk-forward analysis and Monte Carlo simulation?
Walk-forward analysis checks whether a strategy still works on data it was not optimized on. Monte Carlo simulation resamples outcomes to show how fragile or robust the results might be. Both help avoid trusting one lucky backtest.
Can I use TheoryCraft if I trade manually?
Yes. TheoryCraft is useful even if you place trades manually. You can take the ideas you already trade, turn them into testable rules, and see whether they had a real statistical edge across more market data than your journal alone.
What historical market data can I test strategies on?
TheoryCraft provides access to historical market data across 1,600+ instruments. Additional providers and datasets will be added over time.
Which AI providers are available in TheoryCraft?
TheoryCraft currently supports these AI providers: Anthropic, OpenAI, Google Gemini, OpenRouter, Groq, xAI (Grok), Cerebras, DeepSeek, Mistral, Cohere, Fireworks AI, Together AI, Perplexity, Venice AI, NVIDIA NIM, HuggingFace, Alibaba DashScope, Moonshot / Kimi, MiniMax, Claude Code, OpenAI Codex, Qwen (OAuth), GitHub Copilot, Nous Portal, AWS Bedrock, Ollama Cloud.
Do I need to pay for my own server or cloud machine to run backtests?
No. TheoryCraft provides the historical market data and the compute servers needed to run research workflows and backtests in the platform. You do not need to rent a VPS, configure cloud infrastructure, or maintain a separate machine just to test strategies.
Does TheoryCraft connect to my broker or place trades for me?
No. TheoryCraft does not connect to your broker to execute trades, and it does not place orders for you. It is a research and backtesting platform designed to help you validate ideas before you decide what to trade manually or implement elsewhere.
Can I use Claude, Codex, or other AI agents?
Yes. TheoryCraft is being designed to integrate with external AI agents through MCP.
Can I see and export the code generated by TheoryCraft?
Yes. TheoryCraft generates readable research notebooks and strategy code that you can inspect, understand, and export. The goal is to avoid black-box research: you should be able to see exactly what was tested.
Early Launch Offer

Stop guessing. Start proving.

Turn an idea into evidence. Build it, backtest it, and see whether the edge is real before you risk capital.

0d 00h 00m 00s
Claim Launch Offer

Release discount applied at checkout.