Lebs AI

Odds, not hype.

Ask, screen and test ideas in plain English. Every answer shows the steps it ran and the data it used.

Sample data, labelled as such. Education and research, not investment advice.

Buy SPY after a 2% down day, hold 5 days

#	Strategy	Trades	Won	Return
1	Buy after a 2% down day	131	57%	+16.8%
2	Buy and hold SPY	1	—	+17.9%

One sentence in. A tested answer out.

Ask the way you'd say it: test a strategy, scan the market or check a trend. Lebs turns the question into steps it can show you, runs them on the data, and answers plainly, including when an idea didn't beat simply buying and holding.

Numbers in the animation come from Lebs's own backtest engine on its sample history, with 0.05% costs a side. Past results on sample data say nothing about the future.

Ready when you are.

Lebs AI comes ready to scan the market, read your watchlist, check a trend, test an idea and explain a trade, and to say no when you ask it for a tip.

Scan the market

“Which stocks rose most this week, and on what volume?”

Here are the sample symbols that rose most over the last five sessions, with today's volume against their 20-day average.

Biggest movers, 5 sessions
Symbol	Name	5-day	Volume vs avg
NVDA	NVIDIA	+3.1%	1.2×
GOOGL	Alphabet	+2.5%	1.2×
META	Meta Platforms	+2.2%	1.2×
MSFT	Microsoft	+1.4%	0.8×
AMD	Advanced Micro Devices	+1.2%	0.9×

Numbers computed by Lebs from its sample data. Education and research, not investment advice.

Four modes, one box.

Ask

Ask why something moved, what a trade or a level means, or how a concept works. Every answer names its data and when it's from.

“Why did NVDA move today?”

Screen

Describe the stocks you want in plain English. Lebs shows each filter it understood, what it can't answer, then the matches.

“Stocks up 5% this week on twice normal volume”

Backtest

Describe a strategy and Lebs builds the rules for you to check, then runs them on past prices after costs, beside simply buying and holding.

“Buy SPY after a 2% down day, hold 5 days”

Insights

Likely ranges and the chance of finishing higher for five horizons, a goal check, and the model's own track record on unseen data.

“NVDA grades and goal check”

Honest by design.
The model keeps score of itself.

Tested on data it never saw. Every outlook is checked walk-forward: trained only on the past, scored on what came next, and compared with simply guessing the usual direction. When it has no edge, it says so.
Shows its work. Each answer lists the steps it ran and the data it used. Lebs computes every number; nothing is made up by a language model.
Never tells you what to buy. No buy or sell calls and no price targets. Questions asking for one get a plain reply and a link to the lesson that helps you decide for yourself.

Before it answers

Every question is sorted first.

01You askPlain English, up to 500 characters.
02Jev sorts itWhy something moved, something on screen, your journal, a screen, an idea to test, a concept, or advice.
03The right tool answersScreens and tests get their mode; advice requests get the no-advice reply.

Jev is TypeSafe AI's decision model; it sees only the question text, never your name, email or account. If it isn't available, Lebs's own rules do the same, and the answer says which.

Ask better questions.

Try Lebs AI