01 · Once
Connect.
Add AlphaLab as an MCP server in Claude, ChatGPT, Claude Code, Codex or Cursor, and sign in.
The AI quant lab behind your agent.
Connect AlphaLab to the AI agent you already use, like Claude or ChatGPT, and start with a trading hypothesis. Your agent gets the market data, quant tools and research workflows to take it through signal research, backtesting, optimization, robustness and out-of-sample testing.
From hypothesis to verdict.
Your agent does the reasoning. AlphaLab gives it the data, the engine and the research process, from rule-based systematic strategies to advanced quantitative and machine-learning research. A refutation is a result, not a failure.
02 · You
Hypothesize.
Give your agent a trading idea: a rule to stress-test, a parameter to research, or a hypothesis to investigate. You do not need a finished strategy.
03 · Your agent
Evaluate.
It takes the idea through signal research, backtesting, optimization, robustness and out-of-sample testing on AlphaLab. Every run, chart and result is in the browser for you to check. The allocation call stays yours.
Your agent does the work. Every run, chart and result stays in AlphaLab, where you can check it or edit the strategy by hand.
What your agent gets
The lab is the hard part.
An agent is only as good as the tools it can call. AlphaLab gives it data that is already licensed and aligned, an engine that returns the same answer every time, validation it cannot skip, and the full quant methods bench.
Data that is already licensed and aligned.
Stocks, futures, forex, options, prediction markets, and crypto, plus news and fundamentals: contracts paid for, timestamps aligned, survivorship handled, datasets normalized. Your agent starts on clean data instead of writing pipelines.
A deterministic engine.
The same strategy, the same data, the same result, every time. What your agent builds is deterministic code executed on that engine, not a model’s narration of what it might do, so every simulated trade can be reproduced line by line. Reinforcement learning is the one exception, and it is labelled where it is used. Determinism is what makes a result checkable.
Validation that cannot be skipped.
Validation runs on AlphaLab, not inside the agent. It is a deterministic robustness run, never model judgment, so your agent cannot skip it or argue past it. It is part of every plan.
The full methods bench.
Signal extraction, statistical analysis, walk-forward, Monte Carlo, meta-labeling, reinforcement learning, regime testing, leakage checks, parameter stability, and execution testing. Every method is on every plan, and your agent can call any of them.
Full history. Full market. Every asset class.
US Stocks
Equities and ETFs.
US Futures
Index, energy, metals, agriculture.
Crypto
Major venues, 24/7.
Forex
Major and minor pairs.
Options
Equity and index options.
Prediction Markets
Event-driven contracts.
The independent check
Nothing reaches you unchallenged.
Validation runs on AlphaLab, not inside your agent. It is a deterministic robustness run on a platform-enforced configuration, not a second opinion. Before a result is put in front of you, it has to survive four questions, with the rules enforced by the platform rather than improvised by the agent.
Does it hold out of sample?
Strategies are re-fit on rolling windows and judged on the windows that follow. A backtest that holds up on a single split tells you what worked once. Walk-forward tells you whether it keeps working, and by how much it degrades when it stops. The partitioning is enforced at the platform level, not the strategy level, so your agent cannot peek at data it was not given.
Or does it just like one regime?
A result that only works in one volatility regime, one rate environment, or one direction of trend is a result about that period, not about the market. The check separates the two and says which you have.
Does it survive costs?
Edges that live inside the spread are not edges. Fees, slippage, and fill assumptions are stressed, and how much room the strategy has before the cost assumptions matter is reported rather than assumed.
How much can it hold?
Capacity is part of the verdict. A strategy that works at one size and breaks at ten is reported as exactly that, so the allocation call you make is made with the size in front of you.
Then it has to prove it forward.
A strategy that passes can be deployed to a paper account to build a real, forward, out-of-sample record. Walk-forward simulates out-of-sample performance; paper trading measures it. Paper comes before live, and going live is always your call.
Deployment models
Three ways to run it.
01 · Our cloud
Hosted by AlphaLab.
We run it on AWS. You sign in and start.
02 · Your AWS account
Bought in AWS Marketplace.
Subscribe in AWS Marketplace and it deploys into your own account, your VPC, your region, on your existing AWS agreement.
03 · On premise
In your own data center.
The full lab on your own hardware, behind your own firewall.
Same lab in all three. Only the infrastructure changes.
Starting with Claude or GPT is the easy part. The lab behind the model, data contracts, aligned pipelines, the backtesting and execution engine, robustness runs, experiment history, controls, and validation independent enough to be worth having, is years of work before it is a single tested idea.
For a small or medium-sized prop firm, fund, CTA, family office, or emerging manager, the choice is to build that lab yourself, or connect your agent to one that already exists.
Plans.
Firms start with a free scoped pilot and a contract. Individuals start self-serve. Every plan includes the same data, engine, methods and validation, and works with any MCP agent. What changes is how many hypotheses run in parallel.
Starting on Your Desk costs you nothing later: moving up to a Firm Desk carries your hypothesis backlog, your research history and your forward track records across intact.
AlphaLab for firms
Start with a free pilot.
One scoped investigation, about thirty days, taken through the full process to a defended strategy or an honest refutation. That is how a firm relationship starts. From there, a standing desk under your direction, or the Lab Setup program if you have the capital and the theses but no quant capability yet. This is built for small and medium-sized prop firms, funds, CTAs, family offices, and emerging managers. We run a small number of these at a time, so ask about the current slot.
Frequently Asked Questions
How is AlphaLab different from using Claude or ChatGPT on its own?
You keep the agent you already use. It brings the reasoning; AlphaLab brings what it cannot do alone: licensed, aligned market data, a deterministic backtesting engine, the quant methods, validation, and a research history that stays in AlphaLab whichever agent you use next. The allocation call stays yours.
Which agents work with AlphaLab?
Any MCP client, including Claude, ChatGPT, Claude Code, Codex and Cursor. Add the AlphaLab MCP URL to your agent and sign in. Every run your agent starts also shows up in the AlphaLab web app.
Do I need to be a quant?
No. AlphaLab covers the full range, from rule-based strategies, entries, exits, filters and position rules, through statistical and machine-learning approaches. You bring a trading idea worth testing; your agent runs the research process on AlphaLab and can explain any method a result uses. The validation bar is the same across that range.
We already have a quant research team. What's here for us?
Your researchers point their own agents at AlphaLab and skip the layer that eats most research time: data pipelines, normalization, engine work, execution testing, walk-forward configuration, robustness, and validation nobody can accidentally skip. Your researchers keep the judgment, and the capital stays yours.
Why wouldn't I just use Python?
You can. Claude or ChatGPT can write the Python, but you still build and maintain pipelines, normalize datasets, wire backtests and execution, track experiment history, handle multiple-testing risk, and enforce the research process yourself. AlphaLab does that part. If you would rather spend your time on hypotheses than infrastructure, that is what AlphaLab is for.
Is AlphaLab a trading bot or a fully automated trading system?
No. Nothing reaches a live account without your explicit authorization. Strategies go to paper first and build a forward track record; going live is a separate decision you make.
What markets and instruments are supported?
US stocks, US futures, forex, options, prediction markets, and crypto. Full markets across all six, plus news and fundamentals.
Is the data the main differentiation?
No. Licensed data is part of it, but the product is the lab: the engine, the methods, validation, controls, and a research process your agent follows on every idea. The data matters because the research needs it.
Where does AlphaLab run?
Three ways: our hosted AWS cloud, your own AWS account through AWS Marketplace, or on premise in your own data center. Same lab in all three.
How much does AlphaLab cost?
A Firm Desk runs from one researcher to five, with a multi-seat workspace, attribution and sign-off records, and a weekly and monthly cadence; it is contracted rather than checked out, so the researcher count and the terms are agreed with you. Firms start with a free scoped pilot rather than a subscription, and firms with capital and theses but no quant capability yet can take the Lab Setup program. Your Desk is $990/month: one researcher, directed by you, on your own book or on one account before terms. It is the only plan with a list price. Moving from Your Desk to a Firm Desk carries your backlog, research history and forward track records across intact, so nothing is lost by starting small. Already subscribed? Your rate is locked in for as long as you stay subscribed, even when list prices change.