How AI-Powered Crypto Screening Works

Beyond traditional screeners

Traditional crypto screeners give you dropdowns. Pick an indicator, set a threshold, maybe combine two conditions. The result is rigid: you’re limited to what the screener’s UI supports.

vyx takes a fundamentally different approach. Instead of clicking through menus, you describe what you’re looking for. The AI interprets your intent and generates custom filter code that does exactly what you asked for.

Here’s what that looks like in practice.

From description to filter

Say you type this:

“Find pairs where price closed below the lower Bollinger Band on the previous candle and then closed back inside the bands on the current candle, with RSI below 35”

A traditional screener can’t do this. It requires comparing the previous candle to the current candle, checking Bollinger Band position for both, and combining that with an RSI condition. That’s a multi-step, time-aware computation.

vyx’s AI reads your description and generates real code that:

1. Calculates Bollinger Bands (20 period, 2 standard deviations)
2. Checks the previous candle’s close against the lower band
3. Checks the current candle’s close against the lower band
4. Computes RSI(14)
5. Combines all conditions with proper logic

The generated code is actual filter logic, not a preset with adjusted parameters. Every filter is unique to your description.

The scanning pipeline

Once your filter is created, here’s what happens on every candle close:

1. Data collection

vyx pulls the latest OHLCV (open, high, low, close, volume) data for every Binance futures pair. For a 1-hour filter, this happens every hour. For a 5-minute filter, every 5 minutes.

2. Indicator calculation

For each pair, vyx computes the technical indicators your filter needs:

• Moving Averages (MA, EMA) at any period you specify
• RSI with configurable period (default 14)
• MACD with signal line and histogram
• Bollinger Bands with configurable period and deviation
• ATR, Stochastic, VWAP and more

These are calculated fresh from raw price data on every scan cycle, not cached values that might be stale.

3. Filter evaluation

Your filter code runs against each pair’s data. The code has access to the current candle, previous candles, and all computed indicators. When all conditions evaluate to true, a signal fires.

4. Signal delivery

Fired signals include:

• The pair that matched (e.g., BTCUSDT)
• The timeframe (e.g., 1H)
• A chart snapshot showing the exact candle and indicator state when the signal fired
• The filter that triggered it

You see the full context of why the signal fired, not just an alert.

What makes AI generation different

The key insight is that AI doesn’t just fill in parameter values. It generates novel code structures based on your intent.

Consider these two filter descriptions:

Description A: “Find pairs where the 9 EMA crosses above the 21 EMA with RSI above 50 and above-average volume”

Description B: “Pairs where the MACD histogram has been negative for at least 5 bars and just turned positive, with price above the 200 EMA on the daily”

These require completely different code architectures. Description A is a single-point-in-time check. Description B requires iterating over the last 5+ bars to check histogram history, plus referencing a different timeframe (daily) for the EMA condition.

The AI handles both. It understands temporal conditions (“has been… for at least 5 bars”), cross-timeframe references (“on the daily”), and compound logic.

Available data

Every filter can access:

Data Type	Examples
Price (OHLC)	Open, high, low, close for any pair
Volume	Per-candle volume, volume averages
Moving Averages	MA and EMA at any period
Oscillators	RSI, Stochastic (%K, %D), MACD
Volatility	Bollinger Bands, ATR
Volume-weighted	VWAP
Patterns	Engulfing, doji, hammer, etc.

Timeframes range from 1 minute to 1 day: 1m, 3m, 5m, 15m, 30m, 1h, 2h, 4h, 6h, 8h, 12h, 1d.

Speed and coverage

Every scan cycle processes all Binance futures pairs. With hundreds of actively traded pairs and multiple indicator calculations per pair, this is computationally intensive.

vyx handles this by running filter evaluations in parallel and caching indicator calculations where safe to do so. The result: you get signals within seconds of a candle close.

Try it yourself

The best way to understand how it works is to try it:

1. Sign up at vyx.app
2. Browse signals from the built-in filters to see the signal format
3. Upgrade to Pro and describe your first custom filter

You’ll see the AI-generated code before it runs. Full transparency, no black boxes.