Quantrader V1

A 33-Agent Quantitative Trading Platform for Hyperliquid

Built with React + TypeScript + Cloudflare Workers — featuring a 15-agent core engine, 18 AI investor personas powered by Fincept Lab, and a CFA-level quantitative analytics suite.

Inspired by https://github.com/SAHU-01/hyperliquid-ai-trader and https://github.com/Fincept-Corporation/FinceptTerminal

Author: Brian (adefebrian.com)

📋 Table of Contents

Overview
System Architecture
The 33-Agent System
- Core Agents (1–15)
- Fincept Committee (16–33)
Fincept Lab Integration
Dynamic Margin Engine
Model Context Protocol (MCP)
Dashboard Views
API Endpoints
Workflow
Installation & Setup
Configuration
Tech Stack
Project Structure
License

🌟 Overview

This project is a production-grade algorithmic trading system for the Hyperliquid DEX. It combines quantitative finance (GARCH volatility, Kelly criterion, VaR risk management) with multi-agent AI consensus to make autonomous trading decisions in paper or live mode.

Key Highlights

Feature	Description
33-Agent Ensemble	15 core quant agents + 18 AI investor personas (Buffett, Soros, Dalio, etc.) voting together
Fincept Lab	CFA-level analytics, 590+ QuantLib endpoints, macro intelligence, risk stress testing
Dynamic Margin	Position sizing scales from $12 to 60% of NAV based on composite confidence
Fixed Leverage	User-configured leverage (e.g., 20x) stays fixed — never exceeds `max_leverage`
Multi-LLM Support	Alibaba Qwen, OpenAI, Anthropic, DeepSeek, Google Gemini — any OpenAI-compatible API
Paper Mode	Full simulation with realistic fills, fees, TWAP/VWAP slicing, and queue modeling
Real-time Dashboard	6 views: Mission Control, Agent Theater, Fincept Lab, MCP Monitor, Trade History, Settings
Scalping Presets	One-click intensity modes: LOW, MEDIUM, HIGH, AGGRESSIVE
DB Reset from UI	Reset database, portfolio, and settings directly from the dashboard

🏗 System Architecture

┌─────────────────────────────────────────────────────────────────┐
│                     FRONTEND (React + Vite)                     │
│  Mission Control │ Agent Theater │ Fincept Lab │ Settings │ ... │
└─────────────────────────┬───────────────────────────────────────┘
                          │ HTTP / WebSocket
┌─────────────────────────▼───────────────────────────────────────┐
│               CLOUDFLARE WORKER (API Gateway)                   │
│                                                                 │
│  ┌──────────────────────────────────────────────────────────┐   │
│  │              AGENT RUNNER (runCycle)                      │   │
│  │                                                          │   │
│  │  Tier 1: Signal Layer                                    │   │
│  │    ├─ 01 Regime Detection (HMM)                          │   │
│  │    ├─ 02 Alpha Research (IC/IR)                           │   │
│  │    ├─ 03 Sentiment & News (Funding, Fear/Greed)          │   │
│  │    ├─ 04 Microstructure (OFI, Liquidity)                 │   │
│  │    ├─ 05 Volatility (GARCH)                              │   │
│  │    └─ 15 Technical Analysis (MTF Confluence)             │   │
│  │                                                          │   │
│  │  Fincept Committee (18 AI Personas)                      │   │
│  │    ├─ Buffett, Graham, Lynch, Munger, Soros, Dalio...    │   │
│  │    ├─ Fed Analyst, Macro Strategist, Credit Analyst      │   │
│  │    └─ Geopolitics, Energy, China Watcher                 │   │
│  │                                                          │   │
│  │  Tier 2: Quant Engine                                    │   │
│  │    ├─ 06 Statistics (ADF, Kalman)                        │   │
│  │    ├─ 07 Mathematics (Kelly, Cov Shrinkage)              │   │
│  │    ├─ 08 Risk Management (VaR, VETO Power)               │   │
│  │    └─ 09 Portfolio Construction (Fractional Kelly)       │   │
│  │                                                          │   │
│  │  Tier 3: Execution                                       │   │
│  │    ├─ 10 Execution (TWAP/VWAP/Iceberg)                   │   │
│  │    └─ 11 Paper Simulator (Fill + Queue)                  │   │
│  │                                                          │   │
│  │  Tier 4: Feedback Loop                                   │   │
│  │    ├─ 12 Attribution (Brinson)                           │   │
│  │    ├─ 13 Learning (Online Weight Updates)                │   │
│  │    └─ 14 Watchdog (Health Monitor)                       │   │
│  │                                                          │   │
│  │  ► ORCHESTRATOR (Weighted Consensus Voting)              │   │
│  │    → composite > threshold? → agreement ≥ min? → TRADE   │   │
│  └──────────────────────────────────────────────────────────┘   │
│                                                                 │
│  ┌─────────────┐  ┌─────────────┐  ┌────────────────────────┐  │
│  │  D1 Database │  │ MCP Bridge  │  │ Fincept QuantLib Proxy │  │
│  │  (Settings,  │  │ (6 servers) │  │ (api.fincept.in)       │  │
│  │   Trades,    │  │             │  │ 590+ endpoints         │  │
│  │   Portfolio) │  │             │  │                        │  │
│  └─────────────┘  └─────────────┘  └────────────────────────┘  │
└─────────────────────────────────────────────────────────────────┘

🤖 The 33-Agent System

Tier 1–4: Core Agents (1–15)

#	Agent	Role	Key Algorithm
01	Regime Detection	Signal (voting)	Hidden Markov Model, Hurst Exponent
02	Alpha Research	Signal (voting)	Information Coefficient, Ensemble IC
03	Sentiment & News	Signal (voting)	Funding Rate z-score, Fear & Greed Index
04	Microstructure	Signal (voting)	Order Flow Imbalance (OFI), Urgency Score
05	Volatility	Signal (voting)	GARCH(1,1), Realized vs Implied Vol
06	Statistics	Quant Support	ADF Test, Kalman Filter, Cointegration
07	Mathematics	Quant Support	Portfolio Optimization, Covariance Shrinkage
08	Risk Management	VETO Power	VaR (Historical + Cornish-Fisher), Circuit Breakers
09	Portfolio Construction	Sizing	Fractional Kelly Criterion, Dynamic TP/SL
10	Execution	Execution	TWAP, VWAP, Post-Only, Iceberg Strategies
11	Paper Simulator	Simulation	Realistic Fill Engine, Queue Position, Latency
12	Attribution	Feedback	Brinson Model (Allocation + Selection + Interaction)
13	Learning	Feedback	Online EMA Weight Updates, Model Versioning
14	Watchdog	Monitor	Heartbeat Tracking, Latency Anomaly Detection
15	Technical Analysis	Signal (voting)	Multi-Timeframe Confluence (6 TFs), 8+ Patterns

Fincept Strategic Committee: Agents 16–33

These 18 AI personas are powered by your configured LLM (Qwen, GPT-4, Claude, etc.). They run as a single batched API call every 5 minutes and their votes are injected into the Orchestrator alongside core agents.

#	Persona	Category	Philosophy
16	Warren Buffett	Investor	Value investing, economic moats, margin of safety
17	Benjamin Graham	Investor	Deep value, net-net, quantitative screens
18	Peter Lynch	Investor	GARP, tenbaggers, invest in what you know
19	Charlie Munger	Investor	Mental models, inversion, quality at fair price
20	George Soros	Trader	Reflexivity, macro, boom-bust cycles
21	Ray Dalio	Trader	All-weather, risk parity, economic machine
22	Jim Simons	Trader	Quantitative, statistical arbitrage, patterns
23	Stanley Druckenmiller	Trader	Top-down macro, concentration, liquidity flows
24	Seth Klarman	Investor	Margin of safety, capital preservation
25	Howard Marks	Investor	Market cycles, second-level thinking
26	Bill Ackman	Trader	Activist, catalyst-driven, event-driven
27	Cathie Wood	Investor	Disruptive innovation, Wright's Law, S-curves
28	Fed Analyst	Economic	Monetary policy, interest rates, yield curve
29	Macro Strategist	Economic	GDP, inflation, employment, global trade
30	Credit Analyst	Economic	Bond markets, credit spreads, default risk
31	Geopolitics Analyst	Geopolitics	Political risk, sanctions, trade wars
32	Energy Analyst	Geopolitics	Oil/gas, OPEC, mining economics
33	China Watcher	Geopolitics	Chinese economy, PBOC, US-China relations

How it works: The FinceptCommitteeAgent sends a single prompt to the LLM asking it to role-play all 18 personas simultaneously. Each persona returns a direction (+1/−1/0), confidence (0–1), and reasoning. Results are cached for 5 minutes and fed into the Orchestrator's weighted voting system.

🔬 Fincept Lab Integration

The Fincept Lab tab in the dashboard provides four sub-panels:

1. AI Investor Agents

Grid view of all 18 personas with live portfolio context. Click any agent to see their individual analysis using your LLM provider.

2. QuantLib Suite

REST proxy to api.fincept.in — access 590+ quantitative endpoints:

Risk: Value at Risk (VaR), Expected Shortfall, GARCH
Pricing: Black-Scholes, Binomial Trees, Monte Carlo
Portfolio: Markowitz Optimization, Risk Parity
Fixed Income: Bond Pricing, Duration, Yield Curves

3. Market Intelligence

Aggregated real-time market data:

Fear & Greed Index
CoinGecko crypto pricing
Portfolio metrics (NAV, PnL, Win Rate)

4. Risk Analytics

Parametric VaR Calculator
Stress Testing Scenarios
Drawdown Analysis

💰 Dynamic Margin Engine

Position sizing is confidence-adaptive — the system allocates more capital when agents strongly agree, and less when signals are weak.

Composite Confidence → Margin Allocation
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  < 0.30  (Low)       →  ~10% of NAV  ($12 minimum floor)
  0.30–0.60 (Medium)  →  15–30% of NAV
  0.60–0.80 (High)    →  30–45% of NAV
  > 0.80  (Very High) →  45–60% of NAV (capped)

Formula: marginPct = min(0.60, 0.10 + compositeConfidence × 0.625)

Leverage is always fixed at the user-configured value (e.g., 20x). It is hard-clamped to max_leverage as a safety net — the 50x leverage bug has been permanently fixed.

📡 Model Context Protocol (MCP)

The system uses MCP to decouple data acquisition from strategy logic:

Server	Purpose	Default URL
News Sentiment	Real-time crypto news + sentiment	Custom
Orderbook Analyzer	L2 book analytics, imbalance	Custom
Technical Analyzer	20+ indicators (RSI, EMA, BB)	`technical-analyzer-mcp.*.workers.dev`
Signal Generator	Unified alpha signal aggregation	`signal-generator-mcp.*.workers.dev`
Hyperliquid Trader	Exchange API interface	`hyperliquid-trader-mcp.*.workers.dev`
Performance Tracker	Historical PnL + metrics	`performance-tracker-mcp.*.workers.dev`

🖥 Dashboard Views

Tab	Description
Mission Control	Agent consensus bars, open positions, equity curve, risk status, decision log
Agent Theater	Full 33-agent grid with confidence, direction, role, and per-agent details
Fincept Lab	AI personas, QuantLib explorer, market intelligence, risk analytics
MCP Monitor	Server health, latency graphs, ping testing, activity log
Trade History	Complete trade log with entry/exit prices, PnL, hold time, regime
Settings	Capital, leverage, risk limits, LLM config, scalping presets, DB reset

Scalping Intensity Presets

Mode	Leverage	Threshold	Min Agreement	Kelly
LOW (Conservative)	5x	0.30	3	0.10
MEDIUM (Balanced)	10x	0.20	2	0.25
HIGH (HF)	20x	0.15	2	0.50
AGGRESSIVE (Max)	20x	0.10	1	1.00

🔌 API Endpoints

System Control

Method	Path	Description
POST	`/system/start`	Start auto-cycling
POST	`/system/stop`	Stop auto-cycling
POST	`/system/run-cycle`	Run single trading cycle
POST	`/system/emergency-halt`	Emergency halt — close all positions
GET	`/system/status`	System running state

Settings

Method	Path	Description
GET/POST	`/settings`	Get/save system settings (auto-resets portfolio on capital change)
POST	`/settings/mode`	Switch paper/live mode
POST	`/settings/agent`	Toggle/weight individual agents
GET/POST	`/settings/llm`	LLM configuration (API key masked on GET)
POST	`/settings/llm/test`	Test LLM connection
POST	`/settings/reset`	Full database reset

Trading

Method	Path	Description
GET	`/positions`	Open positions (with live mark-to-market)
POST	`/positions/:id/close`	Close specific position
POST	`/paper/open`	Manually open paper position
GET	`/paper/pnl`	NAV, cash, realized PnL, win rate
GET	`/paper/history`	Trade history (last 100)
POST	`/paper/reset`	Reset portfolio to initial capital

Analytics

Method	Path	Description
GET	`/agents/status`	All agent outputs, reasoning, decision
GET	`/equity-curve`	Equity curve data (last 500 points)
GET	`/decision-log`	Orchestrator decision history
GET	`/risk/current`	Current risk metrics
GET	`/performance`	Sharpe, Sortino, win rate, profit factor

Fincept Lab

Method	Path	Description
POST	`/fincept/quantlib/*`	Proxy to QuantLib API (590+ endpoints)
GET	`/fincept/quantlib/*`	Proxy to QuantLib API (GET)
POST	`/fincept/agents/analyze`	Single agent persona analysis
GET	`/fincept/market-intel`	Aggregated market intelligence

MCP Monitor

Method	Path	Description
GET	`/mcp/status`	All MCP server statuses
POST	`/mcp/ping`	Ping specific MCP server
GET	`/mcp/log`	MCP activity log

🔄 Workflow

1. MARKET SNAPSHOT
   └─ MCP Bridge fetches Price, Orderbook, News, Volatility data

2. AGENT ANALYSIS (Every Cycle)
   ├─ 6 Core Voting Agents compute signals (instant, pure math)
   ├─ 18 Fincept Personas provide AI opinions (cached 5 min)
   ├─ Quant Engine calculates risk, sizing, Kelly
   └─ Execution Agent selects optimal strategy

3. ORCHESTRATOR CONSENSUS
   ├─ Weighted sum of all agent signals
   ├─ composite_confidence > threshold? ✓
   ├─ agreement_count ≥ min_agents? ✓
   ├─ Risk Agent VETO check? ✓
   ├─ TA Price Action VETO check? ✓
   └─ Regime + MTF gates passed? ✓ → TRADE APPROVED

4. DYNAMIC MARGIN SIZING
   └─ $12 min ──── scales by confidence ──── 60% NAV max

5. EXECUTION
   ├─ Paper Mode: Realistic fill simulation
   └─ Live Mode: Hyperliquid API execution

6. FEEDBACK & LEARNING
   ├─ Attribution decomposes PnL sources
   ├─ Learning Agent updates signal weights
   └─ Watchdog monitors system health

⚙️ Installation & Setup

Prerequisites

Node.js v18+
Cloudflare Wrangler (global or npx)
An LLM API key (Alibaba Qwen recommended — cheapest for high quality)

1. Clone & Install

git clone https://github.com/Adefebrian/Quantrader.git
cd Quantrader
npm install

2. Backend (Cloudflare Worker)

cd workers/api-gateway
npm install

# Initialize D1 Database
npx wrangler d1 create trading-db

# Local development
npx wrangler dev

# Production deploy
npx wrangler deploy

3. Frontend (React + Vite)

cd frontend
npm install

# Local development (opens on http://localhost:5173)
npm run dev

# Production build
npm run build

4. Configure LLM

Open the dashboard → Settings tab
Select your LLM provider (recommended: Qwen via Alibaba DashScope)
Enter your API key
Click Test Connection → Save

Recommended LLM for cost-efficiency:

Provider: Qwen (Alibaba DashScope)
Model:    qwen-max-2025-01-25
Base URL: https://dashscope-intl.aliyuncs.com/compatible-mode/v1

🛠 Configuration

All configurations are managed via the Settings tab:

Capital & Execution

Initial Capital ($): Starting NAV for paper trading (auto-resets portfolio on change)
Default Leverage: Fixed leverage for all positions (hard-clamped to max)
Consensus Threshold: Minimum composite score to approve a trade
Min Agent Agreement: How many agents must agree on direction
Kelly Fraction: Position sizing aggressiveness (0.1 = conservative, 1.0 = full Kelly)

Risk Limits (%)

DD Daily Halt/Stop: Drawdown thresholds for pausing/stopping
DD Total: Maximum total drawdown before emergency halt
VaR 95% Daily: Value at Risk limit
Max Single/Corr Exposure: Position concentration limits
Max Margin Utilization: How much of NAV can be used as margin

LLM Configuration

Supports any OpenAI-compatible API:

Alibaba Qwen (DashScope)
OpenAI (GPT-4o, GPT-4o-mini)
Anthropic (Claude Sonnet)
Google (Gemini Flash)
DeepSeek Chat
Custom endpoints (Ollama, OpenRouter, etc.)

Database Management

Reset Settings: Restore all settings to defaults
Reset Portfolio: Clear all positions and trade history
Full DB Reset: Nuclear option — wipes everything

🧰 Tech Stack

Layer	Technology
Frontend	React 18 + TypeScript + Vite
Styling	Vanilla CSS (dark theme, glassmorphism)
Backend	Cloudflare Workers (edge computing)
Database	Cloudflare D1 (SQLite at the edge)
Real-time	WebSocket (dashboard events)
AI/LLM	Multi-provider (Qwen, OpenAI, Anthropic, Gemini, DeepSeek)
Quant API	Fincept QuantLib (api.fincept.in)
Data	MCP servers, Hyperliquid API, CoinGecko, Fear & Greed Index

📁 Project Structure

hyperliquid-ai-trader/
├── frontend/                       # React Dashboard
│   └── src/
│       ├── App.tsx                 # Main app (all 6 views)
│       ├── index.css               # Design system + dark theme
│       └── main.tsx                # Entry point
│
├── workers/api-gateway/            # Cloudflare Worker Backend
│   └── src/
│       ├── index.ts                # API Gateway (all endpoints)
│       ├── agents/                 # 15 Core Agent implementations
│       │   ├── agent-01-regime.ts
│       │   ├── agent-02-alpha.ts
│       │   ├── agent-03-sentiment.ts
│       │   ├── agent-04-microstructure.ts
│       │   ├── agent-05-volatility.ts
│       │   ├── agent-06-statistics.ts
│       │   ├── agent-07-math.ts
│       │   ├── agent-08-risk.ts
│       │   ├── agent-09-portfolio.ts
│       │   ├── agent-10-execution.ts
│       │   ├── agent-12-attribution.ts
│       │   ├── agent-15-technical.ts
│       │   ├── agent-runner.ts     # Cycle orchestration
│       │   ├── orchestrator.ts     # Weighted consensus voting
│       │   ├── message-bus.ts      # Inter-agent messaging
│       │   └── types.ts            # All type definitions
│       ├── fincept/                # Fincept Lab Integration
│       │   ├── investor-agents.ts  # 18 persona definitions
│       │   └── fincept-committee.ts # Batched LLM committee agent
│       ├── mcp/                    # MCP Data Bridge
│       │   └── data-bridge.ts      # Market snapshot builder
│       ├── paper/                  # Paper Trading Engine
│       │   └── portfolio.ts        # Position management + PnL
│       └── utils/                  # Math & indicator libraries
│           ├── math.ts             # Statistical functions
│           └── indicators.ts       # Technical indicators
│
├── mcp-servers/                    # MCP Server implementations
├── migrations/                     # D1 database migrations
├── scripts/                        # Utility scripts
├── FinceptTerminal-main/           # Fincept source (reference)
└── package.json

📄 License

This project is licensed under the MIT License — see the LICENSE file for details.

Developed with ❤️ by Brian (adefebrian.com)

Name		Name	Last commit message	Last commit date
Latest commit History 4 Commits
frontend		frontend
hyperliquid-trading-agent		hyperliquid-trading-agent
mcp-servers		mcp-servers
migrations		migrations
scripts		scripts
workers		workers
.gitignore		.gitignore
LICENSE		LICENSE
MCP_URLS.txt		MCP_URLS.txt
README.md		README.md
dashboard.md		dashboard.md
masterplan.md		masterplan.md
package.json		package.json

Folders and files

Latest commit

History

Repository files navigation

Quantrader V1

A 33-Agent Quantitative Trading Platform for Hyperliquid

📋 Table of Contents

🌟 Overview

Key Highlights

🏗 System Architecture

🤖 The 33-Agent System

Tier 1–4: Core Agents (1–15)

Fincept Strategic Committee: Agents 16–33

🔬 Fincept Lab Integration

1. AI Investor Agents

2. QuantLib Suite

3. Market Intelligence

4. Risk Analytics

💰 Dynamic Margin Engine

📡 Model Context Protocol (MCP)

🖥 Dashboard Views

Scalping Intensity Presets

🔌 API Endpoints

System Control

Settings

Trading

Analytics

Fincept Lab

MCP Monitor

🔄 Workflow

⚙️ Installation & Setup

Prerequisites

1. Clone & Install

2. Backend (Cloudflare Worker)

3. Frontend (React + Vite)

4. Configure LLM

🛠 Configuration

Capital & Execution

Risk Limits (%)

LLM Configuration

Database Management

🧰 Tech Stack

📁 Project Structure

📄 License

About

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages