Most signal bots shout BUY or SELL and vanish. AEGIS gives you the complete picture — which gates fired, what the model weighted, and in plain language, exactly why the signal exists.
Every signal shows its reasoning. Every closed trade — wins and losses — are public and unedited. AEGIS is provided for educational market research and transparency only; it is not investment advice or a financial product.
No credit card required · Full access from day one · Cancel anytime
BTC holding firm support at $67,100. RSI recovering from oversold at 43 with volume 1.4× average. Model assigns 78% probability of clearing TP1.
Most signal bots run one universal model for every coin. AEGIS runs fifty-eight — one trained per token, on that token's own history and behaviour.
BTC and FLOKI have different volatility profiles, different liquidity regimes, different participant behaviours. A model trained on BTC data is the wrong tool for a micro-cap. AEGIS trains a separate pair of XGBoost models for each token it monitors — one for the BUY side, one for the SELL side — on that token's own 7,000+ hours of OHLCV history, with its own feature set pruned by SHAP importance. Each token gets the models it deserves.
Most models learn to follow the trend, because most bars in a trend resolve with it. AEGIS trains for the opposite job — the reversal. Training deliberately over-weights the setups the engine exists to trade: price at a structural extreme, momentum exhausted, a genuine candlestick reversal pattern printed. On top of that sits a calibrated confidence gate: it converts the model's raw output into a probability that this specific direction call is correct, and only fires above a threshold chosen from statistical evidence — not a number someone liked.
A model doesn't go live because it trained without errors — it goes live because its out-of-sample record proves it. Before deployment, every model is tested once on data it has never seen, its overlapping trades are collapsed into genuinely independent events, and the statistical lower bound of its directional precision must clear 60% with positive expectancy after fees. A lucky streak on a small sample cannot pass that bar. Models that fail don't fire — they're benched until a retrain earns it.
Every signal surfaces four pieces of information most tools omit: the exact entry price, a five-rung take-profit ladder priced in percent of entry (+0.5%, +1.5%, +2%, +3%, +3.5%), a stop-loss anchored to real market structure, and a plain-English explanation of every factor the model weighted.
No black box. AEGIS shows its work — the same numbers the model processed, in language that doesn't require a quant background to read.
Start Free Trial — No Card RequiredEvery signal includes
BTC is holding firm support at $67,100 — a level respected three times in the past 72 hours.
RSI (14)RSI — Relative Strength Index
A 0–100 momentum gauge. Below 30 = oversold. Above 70 = overbought. 14 = looks back 14 candles.
at 43 is recovering from oversold without being extended. Volume on this candle is 1.4× the 20-bar average, confirming buyers are stepping in. The model assigns a 78% probability of clearing TP1 given current structure. Stop is placed below the support clusterSupport
A price level where buying has historically been strong enough to stop a decline. Think of it as a floor price has bounced off before. at $65,625 — a clean, structure-based exit if the bounce fails.
The model decides the direction; the market decides the timing. A signal fires only when three hard conditions meet: the model's calibrated confidence clears its evidence-set threshold, price is at a tested support or resistance level (never mid-range, never into open air), and a genuine candlestick reversal pattern has printed on the lower timeframe. Everything else — regime alignment, higher-timeframe trend, confluence, conviction depth — no longer silently kills a signal: it grades one. Each check that disagrees is stamped on the signal as a named warning and caps its tier at RISKY, so you always see both the trade and its doubts. Capital-safety vetoes still block outright: a drifting model, a dead or illiquid market, a reward-to-risk below the floor, or portfolio over-concentration.
Before any signal is evaluated, AEGIS classifies the market regime — trending, ranging, volatile, accumulation, or distribution — using a rule-based classifier sharpened by a probabilistic HMM layer. The engine's home turf is the reversal at exhaustion: buying oversold weakness into support, selling overbought strength into resistance. Regime context then grades every signal — a with-trend entry or one fighting the higher-timeframe tide is tagged and capped at RISKY, and liquidity traps fire nothing at all.
An XGBoost model trained on 70+ structural features — OHLCV, EMA slopes, RSI momentum, volume Z-score, ATR normalisation, VWAP deviation, funding-rate alignment, and more — produces a directional probability for each token. That read is one weighted input into the score, not the whole verdict: a confident model still can't fire a trade the market context won't back. Its weight is trimmed automatically when a model's live win rate drifts below its training benchmark, and a critically degraded model is vetoed outright.
Location matters as much as direction. AEGIS only takes entries at structure — a BUY at support, a SELL at resistance — never mid-range. The support/resistance gate scores the level's quality: how close price sits to a real level, whether a recent break has actually retested and held (old resistance becoming support, never a run-up into the level), higher-timeframe 4h/1d confluence, and whether there's genuine room to a sensible target. A weak level, a mid-range setup, an unretested break, or too little reward for the risk is vetoed outright — no valid structure, no trade.
Context earns weight, it isn't just flagged. ADX strength, volume confirmation, regime confidence, macro-bias alignment, fake-breakout heuristics, RSI exhaustion and signal stability each add their own weight to the score — no BUY at momentum peaks, no SELL at capitulation lows. A confirmation layer reads the 1-hour chart for Break-of-Structure / Change-of-Character, RSI and MACD divergence, and volume climax + absorption. When these point against the trade, the score tilts to HOLD. Every factor that moved the score is shown to you, gate by gate.
The final check is economic: reward-to-risk must be at least 1.6:1 after the round-trip cost of the trade, measured to a real structural objective rather than a projected one. Because every signal is an independent recommendation — not positions in one shared book — AEGIS surfaces opportunities across all 60 tokens at once, with each token holding at most one open signal at a time. Stop-loss and five take-profit levels are then set from real current ATRATR-scaled targets
Exit levels are set by ATR multiples, not fixed percentages — so targets adapt to real current volatility, not historical averages., with the trailing take-profit compressed in high volatility so it's reached within a realistic swing instead of round-tripping to break-even.
The training fleet is bigger than what ships. A model goes live only when its untouched-holdout evidence clears the bar — a statistical lower bound of at least 60% directional precision on independent trades, with positive expectancy after fees. Models that can't prove it are removed from the fleet entirely until a retrain earns it back, and tokens with no real Binance USD-M perpetual are benched to monitor-only rather than served from a different market. 59 of the 60 are cleared to fire. SHIB is benched to monitor-only — its perpetual lists as 1000SHIB, so it could only be served from spot.
No signal service publicly admits this. We do — because the track record is public and every underperforming model would show up in it anyway.
All 60 tokens on a single screen. Each card shows current signal state, gate score, regime, RSI, and time since last event. Refreshes every 5 minutes from the live engine.
Gate-by-gate audit trail for every signal: the BUY / SELL / HOLD scores, each gate's weighted contribution, regime, structure quality, momentum & volume — each with the raw value that determined it, plus any hard veto that blocked the trade.
Real-time OHLCV charts powered by the Binance WebSocket feed. All 15 timeframes from 1m to 1M. Up to 60,000 historical candles. Same feed professional platforms use.
Raw XGBoost model probabilities, calibrated win rate, directional confidence, and TP1/TP2/TP3 hit-probability bars. You see the same numbers the model sees.
Live PnL banner on any chart page where AEGIS has an active position: current price vs entry, unrealised return, TP levels hit so far, and time elapsed since signal.
Detected market regime, institutional order flow indicators, macro BTC/ETH context — the same inputs that gate or block the signal engine, visible to you in real time.
All dashboard features are included in the 3-day trial — no credit card, no partial access. Full cockpit, live charts, signal forensics, and real-time data for all 60 tokens from day one.
Quant methods the desks use. Retail price point. Start free — no card, no partial access.
All prices in US Dollars (USD) · Billed monthly · Cancel anytime · Secure checkout
AEGIS is production software, not vaporware. Here is an unfiltered view of what it does well, what it doesn't, and what's being built next.
