ORACLE 🔮

AI-powered market intelligence agent — real-time on-chain data, news, and social sentiment fused into auditable trading signals via a 4-agent reasoning pipeline.

---

The Problem

Crypto markets generate terabytes of signal daily — on-chain whale movements, DEX liquidity shifts, social sentiment spikes, macro news. Human traders can process maybe 1% of it. Existing tools give you raw data dashboards. None of them reason about what it means.

ORACLE closes that gap: it doesn't just aggregate data, it thinks through it using a multi-agent adversarial reasoning chain, then delivers a single actionable signal with a full audit trail.

---

Architecture

┌─────────────────────────────────────────────────────────────┐
│                        ORACLE Pipeline                       │
│                                                             │
│  ┌───────────┐    ┌───────────┐    ┌───────────┐    ┌─────┐│
│  │ Collector │───▶│  Analyst  │───▶│ Validator │───▶│ SIG ││
│  │  Agent    │    │  Agent    │    │  Agent    │    │ GEN ││
│  │           │    │ (MiMo 7B) │    │(Claude S4)│    │     ││
│  │ • On-chain│    │           │    │           │    │BUY/ ││
│  │ • News    │    │ Chain-of- │    │Adversarial│    │SELL/││
│  │ • Sentiment    │ thought   │    │challenge  │    │HOLD ││
│  └───────────┘    │ reasoning │    │+ confidence    └─────┘│
│                   └───────────┘    └───────────┘           │
│                                                             │
│  Every step logged to ReasoningTrace → full audit trail     │
└─────────────────────────────────────────────────────────────┘

Agent Roles

Agent	Model	Role
DataCollector	—	Fetches on-chain (CoinGecko, DeFiLlama), news (RSS), sentiment (X/Twitter)
MarketAnalyst	MiMo 7B	Long chain-of-thought reasoning over raw data, pattern detection
SignalValidator	Claude Sonnet 4	Adversarially challenges Analyst — recency bias, correlation≠causation, black swan
SignalGenerator	—	Synthesizes outputs → BUY/SELL/HOLD + confidence + entry/exit targets

---

Features

• Multi-agent adversarial pipeline — Analyst proposes, Validator challenges, Generator decides
• Auditable reasoning traces — every inference step logged with timestamps, exportable to JSON/Markdown
• Real-time data fusion — on-chain metrics + news + social sentiment in one context window
• MiMo 7B long reasoning — extended chain-of-thought for complex market pattern analysis
• FastAPI server — REST endpoints for integration with trading bots, dashboards, alerts
• Redis caching — market data cached with TTL to minimize API calls
• Docker-ready — single docker-compose up to run full stack

---

Sample Output

{
  "signal": "BUY",
  "asset": "ETH",
  "confidence": 78,
  "risk_level": "MEDIUM",
  "entry_target": 3420.00,
  "exit_target": 3850.00,
  "stop_loss": 3200.00,
  "reasoning_summary": "Whale accumulation detected (+$240M in 6h), funding rates neutral, sentiment turning positive post-ETF news. Validator flagged macro risk (Fed meeting in 48h) — confidence adjusted from 85 to 78.",
  "trace_id": "oracle-20260522-a3f9b2",
  "generated_at": "2026-05-22T14:30:00Z"
}

---

Quick Start

# Clone
git clone https://github.com/Sicanbt/ORACLE.git
cd ORACLE

# Install
pip install -e ".[dev]"

# Configure
cp .env.example .env
# Edit .env with your API keys

# Run single analysis
oracle analyze --asset ETH --timeframe 4h

# Start API server
oracle serve --port 8000

# Docker
docker-compose up -d

---

API Endpoints

POST /analyze          — Run full pipeline for an asset
GET  /signals          — List recent signals
GET  /trace/{id}       — Full reasoning trace for a signal
GET  /health           — Health check

---

Environment Variables

Variable	Description
MIMO_API_KEY	MiMo 7B API key
ANTHROPIC_API_KEY	Claude Sonnet 4 API key
COINGECKO_API_KEY	CoinGecko Pro API key
REDIS_URL	Redis connection URL

---

License

MIT © 2026 Sicanbt