TheQuantGPT
Systematic research at agent speed — on a server you control.
Cursor agents build rule-based backtests with OOS discipline and audit-ready runs/<id>/ artifacts. We maintain the workflow layer.
The gap
AI makes backtests fast. It does not make them trustworthy.
Ad-hoc chats skip out-of-sample splits, hide lookahead, and leave no durable trail. TheQuantGPT Cursor Lab is a research OS — not another chatbot.
How it works
Your server. Your data. Our rails.
A thin wrapper repo on your cloud instance. Cursor agents follow a written constitution. TheQuantGPT MCP tools guide, validate, and specify robustness tests.
-
01
Bootstrap the lab
Clone the wrapper, wire Cursor MCP, drop OHLCV under
data/. Equities, crypto, FX, bonds, metals — any OHLCV asset class. -
02
Prompt a baseline
Agents scaffold
runs/<id>/, write backtest code, report IS and OOS separately, save charts and metrics. -
03
Stress, then package
One robustness test per turn — PSA, cost stress, Monte Carlo. Validated runs bundle to
reports/.
Ownership
You keep the lab. We keep the workflow layer.
- You own
- Server · data · generated code · charts · reports
- We maintain
- Validators · PSA catalog · MCP tools · templates · updates
- Out of scope
- Live trading · performance guarantees · investment advice
Proof
Real agent outputs — not mockups.
These runs demonstrate workflow discipline (OOS splits, saved code, PSA), not investment performance. Sometimes the verdict is “no edge.”
Pricing
Start with a free trial. Convert when the workflow sticks.
You pay separately for Cursor, your cloud server, and LLM API usage.
7-day trial
Full workflow + MCP access on your server
Free
Beta · design partner
3-month minimum · early adopters shaping the product
€1,000 setup
+ €250 / month
Standard Lab
Ongoing research sessions · async support
€3,500 setup
+ €500 / month
Next step
Book a 15-minute demo — or start the trial today.
Tell us your asset class and whether you already have a Linux server and Cursor. We’ll send the bootstrap SOP and API key.