Autonomous Agents and Their Roles

Each agent in the xFractal Agentic System (XAS) is a domain-specialized cognitive unit engineered to process a defined class of inputs, perform contextual reasoning, and generate actionable outputs. Agents are modular, interoperable, and optimized for both real-time responsiveness and task isolation. They are embedded with retrieval mechanisms, scoring logic access, and inference stacks tailored to their specific function.

Below is a detailed overview of each agent, categorized by domain and current operational scope.

---

Aya – Orchestration, Reasoning, Intent Routing

Function: Aya is the meta-intelligent layer of xFractal. She is responsible for prompt interpretation, memory contextualization, agent orchestration, and recursive reasoning.

Architecture:

Built on DeepHermes 3 from Nous Research, Aya incorporates:

• Vector memory: FAISS + MongoDB for semantic recall

• Recursive reasoning: Fractal Consciousness Layer Prompting (FCLP)

• Causality support: Integrated Causal Inference Engine

• Cross-domain synthesis: Hyperdimensional Thought Generator for multi-domain cognitive interpolation

• System observability: OpenTelemetry and Prometheus integration

• Security: OML fingerprinting for traceability and model integrity

Aya ensures that all downstream agents are coordinated within a unified semantic context, enabling accurate, explainable system behavior.

---

Echo – Social Dynamics & Narrative Intelligence

Function: Echo is the agent responsible for evaluating social sentiment, influencer signals, narrative velocity, and propagation dynamics across public and closed networks.

Data Ingestion:

• Sources: Twitter/X (via ElfaAI, TweetScout, TwitterAPI, Masa Network), Telegram (via internal scraper pipelines), Dexscreener, Pumpfun. P.s. Discord, Reddit and Farcaster coming soon.

• Enhancement Layers: LLM-powered sentiment classifiers and amplification estimators trained on Solana-specific corpora. Advanced Natural Language Processing Capabilities.

• Metadata classification: Influencer segmentation, VC/project affiliation via SaharaAI

Inference Outputs: Echo calculates narrative density, influencer amplification, and community reaction gradients, contributing directly to the Hype Score. It generates both time-decayed sentiment scores and directional momentum vectors for each token.

---

Mobu – On-Chain Behavior and Wallet Analytics

Function: Mobu is responsible for ingesting, interpreting, and correlating transactional data, wallet movements, liquidity flows, and token issuance behavior.

Data Sources:

• Indexers: Mobula, Helius, Moralis, Birdeye.

• Execution Metrics: Transaction volumes, LP provisioning, supply distribution, program interactions

• Wallet Tagging: Whale detection, LP concentration, sniper identification, historical P&L trace

Infrastructure:

• Query Layer: GraphQL + REST API composite calls

• Reasoning Engine: Powered by ASI1 LLM from FetchAI for on-chain language interpretability

• Storage: Layered SQL and vector embedding system for trace matching

Mobu contributes directly to Trading Score and provides enriched on-chain intelligence for predictive and due diligence workflows.

---

Sentra – Risk Evaluation & Contract Intelligence

Function: Sentra analyzes token and contract safety using both static attributes and behavioral patterns, identifying red flags such as honeypots, mint manipulation, insider bundling, and historical rug patterns.

Inputs:

• Data sources: Webacy, GoPlus, Rugcheck, Birdeye

• Detection Domains:

  • Contract mutability and mintability

  • LP bundling and dev wallet control

  • Insider allocations and recent rug association

  • Risk simulations based on dev behavior patterns

Architecture:

• Deterministic classifiers for flag generation

• Behavioral pattern matchers for rug risk quantification

• Continuous learning via exploit dataset training

Sentra directly contributes to the Safety Score and serves as a gating layer for execution-related prompts.

---

Vega – Execution and Transaction Intelligence

Function: Vega translates qualified user intent into executable on-chain actions. It handles direct token swaps (DCA, limit orders coming soon).

Capabilities:

• Execution Support: Raydium, Jupiter, Meteora, BelieveApp, Pumpswap, Moonshot

• Security: Hosted in Phala Network’s Confidential Virtual Machines (CVMs) to ensure inference privacy and secure execution

• Simulation: Leverages Birdeye for transaction simulation and preview

• Autonomy Extensions: Mantis DISE integration for identity routing and agent-owned liquidity

Architecture:

• Execution stack with retry logic, simulation checks, and fallback routing

• Multi-agent coordination via MCP server for swarm trading operations

Vega enables agentic-to-manual execution handoff and will eventually support autonomous trading strategy execution under controlled conditions.

---

Solvion – Solana-Literate Protocol Reasoning

Function: Solvion serves as the reference model for Solana-specific intelligence, ensuring agents reason within the parameters and constraints of Solana’s execution and development environment.

Tech Stack:

• LLMs: Dobby-Unhinged LLaMA 3.3 (SentientAGI), Lumo-70B-Instruct

• Knowledge Base: Aggregates protocol documentation, token standard schemas, project whitepapers, and historical ecosystem data

• SDKs: Live data via Adot SDK, updated on release push

Solvion contextualizes prompts requiring technical or architectural understanding of the Solana chain and its ecosystem.

---

Oura – Technical Analysis and Signal Confluence (In Progress)

Function: Oura processes price charts, overlays indicators, and correlates technical patterns with social sentiment and volatility metrics.

Capabilities:

• TA pattern recognition (MACD, RSI, Bollinger Bands, etc.)

• Market structure classification

• Volatility forecast and reversal detection

• Model Backbone: CNN-augmented ResNet + TA-Lib integration

Oura’s future function will enable agentic confirmation of trade setups based on multi-domain confluence.

---

Nova – Portfolio Analysis and Strategy (In Progress)

Function: Nova will analyze portfolio allocations, track realized vs. unrealized P&L, and suggest rebalancing strategies using both backward and forward modeling.

Features:

• Portfolio indexing

• Wallet and token exposure visualization

• Strategy simulation and indexing

• Agent-consensus optimization proposals

Nova will bridge real-time execution signals with longer-term capital strategy.

---

Bravo – Ecosystem and Macro Sentiment (In Progress)

Function: Bravo acts as a macro-level sentinel, ingesting news, project updates, and ecosystem signals that may affect token trajectories.

Data Sources:

• NewsAPI, SolanaFloor, Heurist MCP "ExalResearch".

• Cross-agent alerting when macro conditions affect micro behavior

Bravo operates as Aya’s external sensory node for non-token-specific information.

---

Lynx – Personalization and Behavioral Modeling (In Progress)

Function: Lynx constructs behavioral profiles for individual users, enabling tailored strategy delivery and agent prioritization based on preferences and risk profiles.

Architecture:

• Giza Memory + Intent modules

• Personality segmentation via historical trades, social data, and query tone

Lynx will support persistent personalization across the agentic ecosystem and enhance long-term user alignment.

Intelligent Memory: xFractal’s Vector Database for Real-Time Market Reasoning

xFractal’s vector database is a specialized data store designed to efficiently index, retrieve, and search high-dimensional embeddings. Raw market, social, and on-chain data are first converted into vector representations using advanced embedding models like OpenAI, Cohere, BGE, and DeepSeek. These vectors are then stored in high-performance databases enabling fast similarity search across billions of data points. When an agent needs to analyze past trends or retrieve insights, it performs a nearest neighbor search, retrieving contextually relevant information to enhance decision-making.

This process plays a crucial role in memory augmentation, adaptive learning, and real-time intelligence processing, allowing agents to store and recall historical market signals, sentiment trends, and past strategies with speed and precision. The vector database also enhances knowledge-based reasoning, helping agents synthesize, predict, and strategize based on structured and unstructured data. By integrating this technology, xFractal ensures that its AI-driven agents can search, recall, and leverage knowledge efficiently, providing faster, smarter, and more context-aware decision-making capabilities in dynamic market environments.