How SignalEye turns a chart into a trade signal

1. Chart capture on TradingView

Every coin SignalEye covers has its own timeframe (for example 4-hour candles) and scan schedule. When a coin is due, the scanner opens the TradingView chart for that exact pair and exchange in a headless browser and takes a high-resolution screenshot.

The chart is drawn with the indicators a technical analyst would normally add, currently RSI and EMA:

• RSI (Relative Strength Index) shows whether momentum is strong, weak or stretched.
• EMA (Exponential Moving Average) shows the trend direction and acts as dynamic support or resistance.

Using a real rendered chart, rather than a list of numbers, means the AI sees the same picture a person would: the shape of the trend, the wicks, the gaps and how price reacts at obvious levels. If you want to learn to read the same chart yourself, see How to read candlestick charts and RSI and EMA in practice.

2. A vision AI model reads the chart

The screenshot is sent to a multimodal (vision) AI model with fixed instructions. It is asked to analyse the trend, market structure (higher highs and lows, or lower ones), support and resistance, volume and the indicators, and then to answer in a strict, machine-readable format:

• whether there is a trade at all. "No clear setup" is a normal and frequent answer;
• the direction: BUY (long) or SELL (short);
• an entry price, a stop loss and one to four take-profit targets, read from the chart's price axis;
• a confidence score from 0 to 100 and a short written reason.

The model is instructed to propose a trade only when the setup is clear and the potential reward to the first target is at least 1.5 times the risk to the stop loss.

The answer is then validated in code, not taken on trust. For a BUY the stop must be below the entry and the targets above it in rising order; for a SELL it is the reverse. Anything malformed is discarded. Setups below the minimum confidence score (currently 65 out of 100) are rejected as well. More on this step in How AI analyzes crypto charts.

3. Kronos forecast confirmation

Chart-reading models are good at recognising patterns, but they can also be confidently wrong. So SignalEye asks a second, very different model for an opinion.

Kronos is an open-source foundation model built specifically for financial candlestick (K-line) data. Its authors pre-trained it on more than 12 billion candlesticks from 45 exchanges. Kronos turns each candle's open, high, low, close and volume into tokens and predicts the next candles, much like a language model predicts the next word.

Because the future is uncertain, SignalEye doesn't ask Kronos for a single prediction. It samples many possible paths for the next 12 candles and measures how many of them end in the same direction. A signal goes out only when the chart reading and the forecast point the same way and at least 70% of the sampled paths agree. When they disagree, nothing is sent.

Kronos is a filter, not an oracle. Its limits are covered honestly in AI forecasting models for price action.

4. Price sanity check against Binance

A vision model reads prices from pixels, and occasionally it misreads the axis. Before a signal is accepted, its entry is compared with the live Binance spot price for the pair. If the entry is more than 3% away from the current price, the signal is rejected. This check applies to Binance pairs, which is where SignalEye's coins are tracked.

5. Cooldowns and one signal per coin

More signals is not better. SignalEye never opens a second signal on a coin while the previous one is still running, and after a signal it waits at least 4 hours before the same coin can produce another one. This keeps the feed focused and stops one choppy market from generating a stream of conflicting calls.

6. Delivery on Telegram

Accepted signals are sent straight to Telegram with the pair, timeframe, direction, entry, stop loss, targets and the model's short reasoning. Signals for coins marked Free go to every user; the others go to Premium subscribers. You can follow them in the chat with @signaleye_bot or in the SignalEye Telegram Mini App.

From time to time the team may also publish a signal entered by hand. Those go through the same tracking and appear in the same public track record.

7. Automatic take-profit and stop-loss tracking

Once a signal is live, SignalEye compares it with the Binance spot price every couple of minutes:

• when a target is reached, you get a message (for example "TP1 reached") with the move from entry, and a reminder that many traders move their stop to the entry price at this point;
• when every target is reached, the signal closes as all targets hit;
• when the stop loss is reached, the signal closes as stopped out, or as TP_n hit, then stop if some targets were reached first;
• if neither happens within 72 hours, the signal expires.

Because prices are sampled rather than streamed tick by tick, a very short wick between two samples can be missed. Your own exchange's fills may also differ slightly from the Binance reference price. Closed signals are published on the track record page.

What SignalEye does not do

• It does not trade for you, connect to your exchange or ask for API keys, passwords or seed phrases.
• It does not know your account size or risk tolerance. Position size is your decision. See Risk management and position sizing.
• It does not guarantee results. Markets can ignore any analysis, human or AI.