Introducing AlphaLab: the agentic quant research platform
AlphaLab is the agentic quant research platform. It is built for working quants and systematic traders outside the big firms: people who want the data, the methods, and the iteration speed of an institutional desk without building the infrastructure themselves.
The core of AlphaLab is a research loop you direct rather than run by hand. You frame a hypothesis, build a strategy, and test it. Research Agents run that same loop around the clock on your instructions, generating candidates, backtesting each one, and reporting back. This post is the overview. The posts that follow will work through each part with real examples.
Research Agents: a quant team that works while you sleep
Research Agents are the center of AlphaLab. You give one a goal, for example a mean-reversion idea on a specific instrument, and it works through the research loop on its own. It builds a strategy on the canvas, runs backtests, reads the results, adjusts, and runs again. It pauses for your approval on the steps that matter, hands work back to you, and reports what it found.
What sets this apart is candor. A research agent tells you what works, and it tells you when nothing does. A run that ends with "this idea has no edge in this period, here are three directions worth trying instead" is a successful run. AlphaLab supports research, not signals. It does not predict the market for you. It helps you find an edge that is genuinely yours, and it stays honest when an idea does not hold up.
Every strategy a research agent builds is a transparent graph. Every node is visible, editable, and runnable. You can open what the agent built, change it, and run it yourself. Nothing is hidden in a black box.
AlphaMind: build and learn in conversation
AlphaMind is the conversational layer of the same system. Where Research Agents work autonomously, AlphaMind works alongside you. It suggests components, wires them together, and explains what a strategy or a result is doing. It proposes edits to your canvas that you review and approve, so the strategy stays yours. AlphaMind and Research Agents both run on Claude, with a router that matches the model to the task at hand.
The Visual Strategy Builder
You build strategies on a visual canvas, connecting components that each do one job: pull in market data, compute an indicator, transform a series, define entry and exit rules, or size and simulate a portfolio. There are more than 100 components, spanning data feeds, technical analysis, statistics, volatility, signal processing, pairs and relative value, risk and performance, microstructure, allocation models, and machine learning. No code is required, and because every strategy is a graph, you can always see exactly what it does.
Rigorous methods, run correctly
AlphaLab runs the methods institutional desks use, and it runs them correctly every time. Walk-forward analysis, out-of-sample testing, meta-labeling, optimization, and portfolio allocation across many tickers are all available, none of them requiring code.
A strategy can be run three ways:
Simple. Run it once over a date range with fixed parameters, and read the equity curve, the trade-by-trade log, and the performance metrics.
Discovery. Search ranges of parameters for the settings that perform best, using methods such as Bayesian optimization, grid search, and random search.
Out-of-sample. Validate with a walk-forward test, optimizing on a training window and testing on a later window the strategy has not seen. This is how you tell a real result from one fit to a single stretch of history.
Institutional-grade data and compute
AlphaLab covers US stocks, futures, forex, and crypto: more than 20,000 instruments, with full history, tick by tick. You can set candles at any interval you choose, from one second upward, or build tick-based bars when you want bars that track market activity rather than the clock.
This rests on partnerships an independent quant could not assemble alone. Market data comes from Databento and Massive. The platform runs on Amazon's cloud infrastructure in the US, so a backtest runs at the size the work requires, not the size your own hardware allows.
What comes next
This was the overview. The posts that follow will go deep, one piece at a time, with real examples: a full research-agent run from goal to result, building a strategy on the canvas, the methods that keep a backtest honest, and the data behind it. We will iterate on real strategies and show the results, including the runs where the answer was no.
If you are a working quant or a systematic trader, the fastest way to understand AlphaLab is to put a research agent to work on an idea you already have. Open AlphaLab and start there.