NVDA Thesis in. Verdict out.
AI trading assistant that researches and backtests your ideas
Describe a thesis or paste a ticker. Agenttrading summarizes the fundamentals, backtests the rule on historical data, and explains the risks in plain English. Educational analysis only. You decide.
18,437+ ideas backtested
Used by swing traders, analysts & investment clubs
- 1 THESIS
- 2 EVIDENCE
- 3 BACKTEST
- 4 RISK
- 5 VERDICT
02 EVIDENCE · FUNDAMENTALS
04 RISK · IN PLAIN ENGLISH
Past performance does not guarantee future results. Educational analysis only, not financial advice.
20+ years of daily datasplits & dividends adjusted
Every assumption shownrange, costs, parameters
Honest verdictslosers stamped in coral
No execution. No promises.analysis only, you decide
01 THESIS - WHAT YOU BRING
Your next idea deserves better than a hunch and a spreadsheet weekend
The manual loop is brutal: a screener, ten filings, a hand-built backtest that silently overfits, and no one to tell you the idea was a regime bet all along. Agenttrading runs that loop properly, in minutes, and shows its work.
You type
"Buy the dip on SPY when RSI drops under 30"
The bench answers
A restated rule, the evidence, a 20-year backtest, the risks, and a verdict you can defend.
Plain English in, backtest out
Type "buy the dip when RSI drops under 30" and Agenttrading extracts the testable rule, shows it to you before running, and backtests it on 20+ years of adjusted daily data. No Pine Script, no logic blocks, no spreadsheet weekend.
the rule is shown before it runsHonest verdicts, even when your idea loses
When a strategy underperforms buy-and-hold, the verdict says UNDERPERFORMED, in coral, on a chart you can read. Tools that sell signals cannot afford that sentence. We can, because analysis is all we sell.
losing verdicts shown by designEvery risk explained before you act on anything
Max drawdown in words, not just a percentage. Concentration, regime dependence, and sample-size warnings on every run. You see what history says could go wrong before capital enters the conversation.
drawdown, concentration, regime, sample sizeHours of research into minutes per idea
The manual loop is a screener, ten filings, and a spreadsheet backtest: 4 to 6 hours per idea. Agenttrading runs the same loop in minutes, with every assumption logged so you can defend the result.
fundamentals + backtest + risk in one pass02 THE BENCH - HOW IT WORKS
Five stations between an idea and a verdict
Every idea moves left to right through the same bench. Nothing is hidden: the rule is shown before it runs, and every assumption is listed with the result.
-
01
Bring a thesis
Describe an idea in plain English or paste a ticker. "Hold NVDA above the 50-day" is enough. Agenttrading extracts a testable rule and shows it to you before anything runs, so you always know exactly what is being tested.
-
02
See the evidence
The assistant summarizes the fundamentals that bear on your idea: revenue trend, margins, valuation, and at least one risk flag. Sources are historical filings and market data, summarized like a patient analyst would.
-
03
Watch the backtest
The rule runs on 20+ years of split- and dividend-adjusted daily data. The chart shows your strategy against buy-and-hold, with the max-drawdown window shaded and every assumption listed: date range, costs, parameters.
-
04
Read the risks
A plain-English risk panel explains what hurt: how deep the drawdown went, how concentrated the bet is, which market regime the rule depends on, and whether the trade count is too small to trust.
-
05
Get the verdict
HELD UP, MIXED, or UNDERPERFORMED, stamped on the analysis with the metrics that back it. The verdict is historical, not predictive. You decide what to do with it; we never execute trades or give advice.
Past performance does not guarantee future results. For educational and informational purposes only. Not financial advice. Consult a licensed advisor.
03 THE MATH - TIME, NOT MAGIC
Hours of vetting become minutes. That is the whole promise.
We will not promise returns; nobody honest can. What we can quantify is process: a manual idea-vet (screener, filings, spreadsheet backtest) runs 4 to 6 hours. The bench does it in minutes, with the assumptions logged.
min
per idea, not evenings
9
assumptions logged per run
20y
of adjusted daily data
Your research hours, backTIME ONLY. NO RETURN MATH.
Manual vetting / month
h
Hours back with the bench
~h
04 FOR TEAM LEADS - ONE PROCESS
For research and fund teams: one vetting process, fully auditable
Every analysis logs its assumptions, parameters, and data range. Junior analysts vet ideas exactly the way seniors do, the debate moves from "whose spreadsheet is right" to "what does the bench say", and risk is on the table before capital is discussed.
- ✓ A shared audit trail of every assumption on every backtest
- ✓ 4 to 6 hours of manual vetting per idea becomes minutes
- ✓ Compliance-friendly framing: educational analysis, never advice
Enterprise signals
SSO / SAML
Roles & seats
Shared audit log
SLA & priority support
Invoicing
Security review & DPA
Worked example, time only
A 4-analyst desk vetting 10 ideas a week at ~5 manual hours each spends ~200 hours a month. The same desk on the bench spends a fraction of that and keeps a written assumptions log for every idea it kills or keeps.
05 THE VERDICT - WHAT USERS SAY
Shorter debates, faster kills, cleaner process
“I killed a bad thesis in 10 minutes instead of losing a weekend to it. The rule looked clever until the bench showed me it was just a leveraged bet on one regime.”
Marcus T.
Swing trader, trades around a day job
“The drawdown chart ended a two-week debate on our desk. Same data, same assumptions, printed on one page. Nobody argues with the assumptions log.”
Priya S.
Senior analyst, 4-person research team
“Our club used to argue from four different spreadsheets. Now we argue from the same backtest, and the meetings got shorter and better.”
Dan K.
Treasurer, 12-member investment club
PRICING - ALL PLANS PAID, ALL HONEST
Less than one bad trade's commission-and-slippage lesson
TrendSpider runs $48 to $228 per month and Trade Ideas $89 to $178. Agenttrading starts at $15 per month, and the verdicts do not flatter you.
Starter
The curious active investor
$15/mo
billed yearly
- 20 analyses per month
- 10 years of daily historical data
- Plain-English thesis to backtest
- Core risk panel on every run
Analyst
Serious swing traders and part-time researchers
$39/mo
billed yearly
- 200 analyses per month
- 20+ years of adjusted daily data
- Options strategy backtesting
- Full risk panel: drawdown, concentration, regime
Pro
Power users and prosumer quants
$103/mo
billed yearly
- Unlimited analyses
- Batch idea testing (run a whole list)
- Custom cost and slippage assumptions
- Priority compute
Enterprise
Research and fund teams, clubs at scale
Custom
- SSO and roles with seats
- Shared, auditable analysis trail
- SLA and priority support
- Invoicing and procurement
Cancel anytime. No execution, no advice, on any plan. Full pricing detail
OBJECTIONS, ANSWERED STRAIGHT
Is AI trading analysis legit? Fair question.
The honest answers, including the ones that say "no". More on the FAQ page.
No, and that protects you. Agenttrading shows educational analysis with every assumption visible; it never tells you what to buy or sell. No one who has not seen your full financial picture should, which is why personalized recommendations are the job of a licensed advisor, not software. Every analysis carries the same two lines: past performance does not guarantee future results, and this is educational analysis only. Consult a licensed advisor before acting on anything.
No. There is no trade execution, no brokerage connection, and no order routing of any kind, by design. Agenttrading is analysis only: it summarizes fundamentals, backtests rules on historical data, and explains risks. You decide what, if anything, to do with the result, in your own brokerage account, on your own judgment.
No, never. Agenttrading does not produce buy lists, signals, or personalized recommendations. It tests YOUR idea and explains what 20+ years of historical data say about it: how the rule performed, what the drawdowns looked like, and where the risks concentrate. The verdict is a statement about history, not a prediction or a recommendation.
No. Past performance does not guarantee future results, and we print that sentence on every chart rather than in a footnote. A backtest tells you how a rule would have behaved in past markets, which is genuinely useful for killing bad ideas and understanding risk, but markets change and rules that worked can stop working. Anyone who promises otherwise is selling something else.
Historical daily prices adjusted for splits and dividends, plus company fundamentals from public filings, covering 20+ years for US-listed stocks and ETFs. Every backtest lists its assumptions with the result: the exact date range, the cost per trade assumed (0.1% by default, configurable on Pro), and the rule parameters. If an assumption changes the answer, you can see it and change it.
No. You describe the idea in plain English or paste a ticker, and the assistant extracts a testable rule, shows it to you before running, and executes the backtest. There is no Pine Script, no block-based logic builder, and no formula syntax. If the assistant misreads your idea, you see it immediately in the restated rule and can rephrase.
Yes. Monthly plans cancel at the end of the billing month and yearly plans at the end of the term, both from your account settings with no calls and no forms. Plans start at $19 per month, which is less than one bad trade's commission-and-slippage lesson, and well under TrendSpider (about $54 to $321 per month) or Trade Ideas ($127 to $254 per month).
Active self-directed investors and swing traders who vet ideas before sizing them, part-time traders with day jobs, research and fund teams that want one consistent, auditable vetting process, and investment clubs that would rather argue from the same backtest than from opinions. It is not for anyone looking for signals to follow blindly or a bot that trades unattended; we do not make one.
From the blog
Backtesting and strategy guides, caveats included
What Is a Good Alpha in Investing? Stocks and Funds
What is a good alpha? Positive, and 1% to 3% a year after fees is genuinely good. Here are the ranges for stocks, funds and portfolios, and why alpha decays.
Is Backtesting Accurate? Why Backtests Overstate Live Results
Is backtesting accurate? It describes the past under stated assumptions, not the future. Here is how much edge decays out of sample, and the six leaks that cause it.
How to Backtest a Momentum Strategy (and Dual Momentum)
How to backtest a momentum strategy: pick the lookback, decide the skip-month, charge real turnover costs, force the test through 2009, and benchmark against buy-and-hold.
The bench is open
Your next idea deserves a verdict, not a hunch.
Bring a thesis. Get the evidence, the backtest, the risks, and an honest stamp. Then decide for yourself.
Past performance does not guarantee future results. For educational and informational purposes only. Not financial advice. Consult a licensed advisor.