AI Agent

How the agent thinks

Every trade signal goes through a 9-step analysis pipeline before a decision is made. Here's exactly what happens — and why.

The 9-step pipeline

1

Market Data

OHLCV candles (up to 500 bars) via CCXT for any pair and timeframe.

2

Indicators

RSI, EMA (9/21/50/200), MACD, Bollinger Bands, ADX — computed from the raw candle data.

3

SMC Analysis

Fair Value Gaps, Order Blocks, Break of Structure, Change of Character, and liquidity sweep detection.

4

Chart Patterns

Head & Shoulders, double tops/bottoms, triangles, flags, wedges — classical price action patterns.

5

Order Book (L2)

Real-time bid/ask imbalance and liquidity walls from the exchange order book.

6

News Sentiment

CryptoPanic + CoinGecko headlines scored for bullish/bearish sentiment weighted by recency.

runs in parallel with step 7
7

On-Chain Data

Funding rates, open interest, liquidation clusters (Coinglass) and whale netflow (Santiment) — run in parallel with news.

runs in parallel with step 6
8

Decision

The agent synthesizes all tool outputs and returns ENTER_LONG, ENTER_SHORT, or HOLD — with full reasoning.

9

Risk Sizing

Position size calculated from account balance, risk %, entry, and stop-loss, with fees and slippage modelled in.

What each tool contributes

Market data + indicators

The agent always starts with raw price data — OHLCV candles sourced from CCXT (the same library used to execute orders). Indicators run on top of those candles. A minimum of 200 candles is fetched to ensure EMA-200 and ADX have enough history to be statistically valid.

SMC & chart patterns

Smart Money Concepts are structural signals — areas where large institutions likely accumulated or distributed. The SMC tool identifies key levels (Order Blocks, FVGs, sweep zones) and saves them to the database so they persist across analysis runs. The pattern tool layers classical formations on top.

Order book depth

L2 order book data reveals real-time buy and sell pressure. Large clusters of limit orders act as support or resistance walls. The agent uses this to confirm or invalidate levels found in SMC analysis.

News & on-chain

News sentiment and on-chain data are fetched in parallel to reduce latency. Elevated funding rates suggest an overleveraged market ripe for a squeeze. Liquidation clusters reveal where stop hunts are likely. Whale netflow signals capital moving into or out of an asset.

Honesty constraint

The agent is explicitly instructed to cite only data returned by its tools — it cannot invent price levels, funding rates, or sentiment scores. If a data source is unavailable, the agent notes this rather than fabricating a value.

Output format

Every decision returns one of three actions:

ENTER_LONG

Bullish confluence — buy signal with TP/SL

ENTER_SHORT

Bearish confluence — sell signal with TP/SL

HOLD

No clear edge — no trade placed

See the agent in action

Open AI Chat and ask it to analyse any coin — no setup needed to start.

Try it free