Search "AI crypto trading signals" and you'll find hundreds of products. Almost none of them are actually running inference on a trained model.
What they're usually running: RSI + MACD + a few moving averages, dressed up in "AI-powered" marketing copy. There's no model. There's no training data. There's no inference. There's a rules engine with a buzzword.
Here's how to tell the difference — and where to find signals that are actually model-driven.
The Red Flags
1. No training data story
Real AI models are trained on data. If a product can't tell you what data the model was trained on, what labels were used, and how it was evaluated — it's not a model. Ask any "AI signal" provider: "What's your training dataset?" If they can't answer in a sentence, walk away.
2. Binary buy/sell with no confidence score
A good model doesn't say "buy." It says "Bullish with 73% confidence." If you're getting binary signals with no uncertainty quantification, you're looking at a rule-based system.
3. Backtested equity curves only
Any system can be tuned to look good in backtest. What matters is out-of-sample performance — how the model does on data it's never seen. If the provider shows you only beautiful backtests with no out-of-sample validation, that's a warning sign.
4. Signals for 500+ tokens
Running a quality inference pipeline at scale costs money. If a free service is covering 500 tokens with "AI signals," the signals are not model-generated — they're programmatic indicator outputs.
What Real AI Signals Look Like
The sovereign-v2 model running on SovereignAI's V100 GPU covers 15 tokens. Not 500. 15.
Why 15? Because running real inference — ingesting price data, volume profiles, cross-token correlations, and bid/ask spread — is computationally meaningful. Quality over coverage.
Each signal output includes:
- Direction: Bullish / Bearish / Neutral
- Confidence: 0–100%, with sub-60% outputs suppressed as Neutral
- Timeframe alignment: 1h, 6h, 24h — all three frames must agree for high-confidence signals
- Volume confirmation: signals on anomalous volume get weighted differently
The model refreshes every 30 seconds. When the inference server is unreachable, the system degrades gracefully to price-only display — it doesn't fabricate signals.
Multi-Timeframe Alignment: Why It Matters
Single-timeframe signals are noisy. A token that's Bullish on 1h might be in a longer-term downtrend on 6h. High-conviction signals require agreement across frames.
SovereignAI weights signals by timeframe coherence:
| Alignment | Signal Quality | |---|---| | 1h + 6h + 24h all agree | High conviction | | 1h + 6h agree, 24h neutral | Moderate | | 1h only | Low conviction → output as Neutral |
This is why the confidence score matters. A 78% Bullish signal on SOL means: the model sees multi-timeframe alignment with strong volume confirmation. A 58% signal means: weak alignment, don't trade on it.
Free vs. Paid Signals
SovereignAI's signals are currently free at /signals. The 15-token grid refreshes live. No account required.
The inference API — which lets developers call GET /api/v1/signal?token=SOL programmatically — will move to a paid tier as the API scales. The free tier will remain for the top 5 tokens.
Revenue from API subscriptions flows into the $SOVAI staking pool — the same pool that receives SovereignSwap trading fees.
How to Use These Signals
A few practical rules:
- Don't trade on a single signal — use signals as one input among many, not as execution triggers
- High confidence + volume confirmation — the most reliable setups are 70%+ confidence signals where DexScreener volume is 2× the 7-day average
- Check timeframe alignment manually — the /signals page shows the direction; combine it with your own 1h and 4h chart read
- Neutral is information — if the model can't find alignment, that's telling you something about the current market regime
Where to Start
- /signals — live 15-token grid, free, no account needed
- API:
GET /api/v1/signal?token=SOL— returns JSON with direction + confidence - In the swap UI — signals appear inline when you select a token pair