> For the complete documentation index, see [llms.txt](https://onex-3.gitbook.io/onex-document/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://onex-3.gitbook.io/onex-document/swap/ai-intelligence.md). # Ai intelligence ## 23. AI Market Regime Detection The system continuously classifies the current market environment using a **Hidden Markov Model** trained on historical price data, volume profiles, funding rates, and on-chain flow metrics. | Market Regime | System Response | | ---------------- | ----------------------------------------------------------------------------------------------------------------- | | Bullish Trending | Momentum signals activated, LP fee buffers adjusted, arbitrage sensitivity increased | | Bearish Trending | Risk warnings elevated, protective option suggestions surfaced, leverage warnings increased | | Sideways / Range | Range-bound LP strategies highlighted, options volatility selling strategies shown (V2) | | High Volatility | Fee buffers increased to protect LPs from impermanent loss, execution splitting recommended | | Liquidity Stress | Spread widening alert, emergency exit paths surfaced. *Note: This indicator may lag the onset of genuine crises.* | *** ## 24. Causal AI Predictive Price Intelligence ### Correlational AI vs. Causal AI A correlational model says: *"ETH and BTC move together."* A causal model says: *"This ETH move is caused by ETH-specific ecosystem news, not BTC — therefore BTC is unlikely to follow."* This distinction is critical for trade quality. ### Technical Methodology | Component | Technology | | ---------------------- | ------------------------------------------------------------------------------------------ | | Causal Inference Layer | DoWhy library implementing structural causal models | | Pattern Recognition | LSTM Neural Networks for historical price context | | Order Flow | Real-time demand pressure using VPIN (Volume-Synchronized Probability of Informed Trading) | | Cross-Asset Graph | Multi-token causal relationship modeling — updated weekly | ### Prediction Accuracy Tracker — Transparency Feature Live dashboard showing historical accuracy of all ONEx AI predictions: * Directional accuracy % * Mean absolute error * Percentage of actual prices falling within predicted range * Updated daily with 30-day rolling window * Full historical data downloadable > *If the model is genuinely good, publishing real accuracy data is our strongest competitive advantage. If it is not good, this forces improvements before users rely on it.* *** ## 25. On-Chain Verifiable AI (ZK-Proven Inference) ONEx is the first DEX to make its AI decisions **independently verifiable** using Zero-Knowledge Proofs. When the AI makes a recommendation or executes an automated action, it generates a ZK proof that any user can verify without revealing the model weights. ### What the ZK Proof Guarantees * The recommendation was made by model version X (hash verified on-chain) * The inputs used were exactly these market signals at this timestamp * The model weights have not been altered since the last governance-approved update * The output was not tampered with between inference and display **Technical implementation:** RISC Zero or SP1 zkVM for ZK-verifiable execution of AI inference. On-Chain AI Model Registry stores model hashes in a governance-controlled smart contract. *** ## 26. Explainable AI Dashboard Every AI decision includes complete, transparent reasoning visible to the user. Uses **SHAP (SHapley Additive exPlanations)** values for feature attribution — showing exactly which signals contributed to each decision and by how much. ### Capital Movement Thresholds — Safety Governance | Threshold | Requirement | | ------------------ | ----------------------------------------------------------- | | Up to $50,000 | AI autonomous action (within user-approved strategies) | | $50,000–$500,000 | Requires explicit user approval per action | | Above $500,000 | Requires multi-sig DAO governance approval | | Emergency override | Users can pause all AI automation instantly from any device | *** ## 27. Behavioral Finance AI + Tilt Detection ### Cognitive Bias Detection Library | Bias | Detection Method | | ----------------- | ------------------------------------------------------------------------------------------------------------------------------ | | Confirmation Bias | One-sided research patterns — only reading bullish or bearish signals | | Anchoring | Pricing decisions anchored to irrelevant reference points (e.g., purchase cost) | | Sunk Cost Fallacy | Holding losing positions disproportionately longer than winners | | FOMO Trading | Entering positions within 30 minutes of sharp rallies without planned entry | | Recency Bias | Overweighting most recent outcomes in strategy decisions | | Tilt Detection | Sequence modeling detects revenge trading after losses — escalating position sizes, shortened time between trades after a loss | > **PRIVACY GUARANTEE:** All behavioral analysis is performed locally in the user's browser. Raw trading behavior data is never uploaded to ONEx servers without explicit opt-in consent. *** ## 28. Federated Learning Behavioral Engine ONEx uses **Federated Learning** to improve its behavioral AI models without ever accessing raw user trading data. Only mathematical gradient updates (not data) are aggregated. This is the only privacy-compliant approach to training behavioral AI on user data at scale. *** ## 29–35. Additional AI Intelligence Features | Feature | Technical Specification | | ------------------------------------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Social Sentiment Signals** | Twitter/X (Enterprise API), Reddit, Discord, Telegram. NLP pipeline with 5–30 second latency. Used for 1-hour+ predictions only. NOT connected to automated execution. | | **AI Narrative & Sector Rotation Tracker** | Monitors narrative momentum across DeFi sectors (AI tokens, DePIN, RWA, L2, DeFi Blue Chips) using on-chain capital flow + social sentiment. Informational only. | | **AI Portfolio Immune System** | Detects stablecoin depegs (price below $0.995 on 3+ oracle sources), known exploit patterns, concentration risk when >30% in a single protocol. Issues alerts — does not act autonomously. | | **Personalized AI Trading Assistant** | Natural language interface for informational queries and strategy exploration. Voice commands limited to informational use — not trade execution (2–5 second processing latency creates price execution risk). | | **Reinforcement Learning Market Making** | RL agent learns optimal fee schedules from historical LP return data. Updated weekly via governance vote. Provides better LP returns across all market regimes than static parameters. | | **Autonomous LP AI Agents** | Auto-rebalances LP position ranges, exits during extreme volatility events. Built on proven infrastructure (Gamma Strategies / Arrakis Finance model). Users set risk parameters; agent operates within them. | | **Governance Assistant AI** | Summarizes governance proposals, analyzes economic impact, surfaces relevant historical precedents. Advisory only — cannot vote or submit proposals autonomously. |