An MCP server for AI agents
Claude Code made models useful for people who already write software. Lagias does the same for people who already trade.
Research born at
How it works
“Trend-following on tech, but I hate drawdowns.” You already trade. The backtest stack is the part you should not have to build.
The same factor data and the same backtest, every time. Claude, ChatGPT, Cursor, or the Lagias app. You talk; the harness runs the numbers.
Whether the backtest is overfit, in plain English. You decide what to do with it.
What you get
Roughly 244 equity factors, re-researched and refreshed every week. Your AI reads current values by ticker. You do not have to maintain the book.
Backtests and walk-forwards run identically for everyone. Ask twice and get the same answer, to the digit.
Deflated Sharpe and the probability the backtest is overfit, graded by code the model cannot see. You decide what to do with it.
The referee
The model never grades its own work. Deterministic code on data it never saw, with the same overfitting checks a quant fund uses. Your AI cannot talk the numbers up, and neither can we.
You talk to Claude, ChatGPT, Cursor, or the Lagias app. The idea is yours. The test runs here.
Deterministic, overfitting-aware scoring on unseen data.
The verdict comes back with the caveats attached. The call is yours.
Example verdicts
The edge survives out-of-sample and through walk-forward. This one earns its place.
Sharpe deflates from 2.4 to 0.2 once we account for how many variants were tried. Don’t risk real money on it.
Both answers ship in the same format, so you or your AI can read either. Figures are illustrative.
Conflicts, disclosed
We charge a flat software fee, the same whether the referee says robust or likely overfit. A broker earns when you trade and a portfolio manager earns on the assets held, so both have a reason to want an encouraging answer.
Go deeper
Everything above rests on statistics you can check without taking our word for it. The handbook explains them in plain language, including the places where the honest answer is unhelpful. The blog argues with the industry, starting with our own product.
Start free
Create an account and test it. You already trade. This is the harness that runs the backtest and tells you whether the result is overfit.
The free tier needs no payment details. Backtested results are not future performance; every verdict quantifies overfitting risk.
FAQ
Lagias is a quant research harness for people who already trade and want to use AI on their ideas. Claude Code made models useful for people who already write software. Lagias does the same for trading: you bring the idea, it runs the backtest, a referee says whether the result is overfit.
Traders who already have a view and want AI to help test it. Family offices and wealth managers later. Not the person who already runs their own backtests and hunts alpha with a native stack. You should not have to build that desk to get an honest grade.
Weekly-refreshed equity factor values, deterministic backtests, and a verdict: deflated Sharpe and the probability the backtest is overfit. You can talk in plain language. The grading step never involves a language model.
Language models are unreliable judges of their own output. 62% of US retail investors already use AI to inform investment decisions, yet only 23% mostly or completely trust its output (Investing.com survey, April 2026). A deterministic pipeline the model cannot game is how we close that gap.
Four checks run on every candidate. Deflated performance, probability of backtest overfitting, out-of-sample testing, and walk-forward validation. Grading always happens on data the proposing model never saw.
No. Lagias is a research and education tool, not a broker. It will not place orders, it returns no buy or sell instructions, and it does not give personalized investment advice. You make every decision.
Now. Create an account at https://app.lagias.com. The free tier needs no payment details. How to add Lagias as a connector, or with a key, is at https://lagias.com/agents.
