Redhound Internal Documentation¶

Enterprise-grade multi-agent AI system for intelligent market analysis and trading decisions.

Python 3.12 LangGraph Docker Ready

Quick Start¶

Installation

Get started in under 5 minutes with uv package manager

git clone https://github.com/redhound-labs/redhound.git
cd redhound
make quickstart

Developer Onboarding

Configuration

Configure LLM providers and data sources with environment variables
```
export OPENAI_API_KEY=your_key_here
export ALPHA_VANTAGE_API_KEY=your_key_here
redhound
```
Config Reference
Mock Mode

Cost-free development and testing without LLM API calls
```
export REDHOUND_MOCK_MODE=true
redhound
```
Mock Mode Guide

Docker Deployment

Production-ready containerized deployment with all services

make up-build   # build and start stack
curl http://localhost:8000/health
docker exec -it redhound-app redhound

Docker Setup

Key Features¶

Multi-Agent Collaboration

A team of 10 specialist analysts collaborate through LangGraph workflows with built-in memory and state management.

Confidence-Gated Debate

Bull and bear researchers engage in structured debates to refine investment theses. A dynamic confidence gate skips the debate when signals are already high-confidence, optimizing for speed and cost.

Vector Memory System

pgvector-powered memory enables agents to learn from past decisions and retrieve relevant historical context for current market situations.

Deterministic Risk Overlay

A 100% deterministic risk management engine applies precise position sizing and stop-loss levels based on market volatility regime and signal confidence.

Signal Accuracy System

Entropy-based confidence scoring, VIX linear interpolation, RSI divergence detection, OBV confirmation, multi-timeframe alignment, sector rotation, and adaptive analyst weights that improve automatically over time.

Background Scanner & Screener

Continuous market scanning with configurable conditions, deterministic quick-screen (zero LLM), and automatic full-analysis for high-scoring opportunities.

System Architecture¶

The project employs a modular monolith architecture with a clear separation between the LangGraph orchestration layer, specialized agents, and a unified service layer.

graph TB
    subgraph "Core Analysts (Parallel)"
        A1[Technical Analyst]
        A2[Fundamentals Analyst]
        A3[Sentiment Analyst]
        A4[News Analyst]
        A5[Market Context Analyst]
    end

    subgraph "Edge Analysts (Optional)"
        A6[Sector Analyst]
        A7[Insider Analyst]
        A8[Options Flow Analyst]
        A9[Short Interest Analyst]
        A10[Earnings Revisions Analyst]
    end

    subgraph "Signal Processing"
        SA[Signal Aggregator]
        CG{Confidence Gate}
    end

    subgraph "Research Team (Debate)"
        R1[Bull Researcher]
        R2[Bear Researcher]
        R3[Research Manager]
    end

    subgraph "Risk Strategy"
        D4[Risk Overlay]
    end

    A1 & A2 & A3 & A4 & A5 --> SA
    A6 & A7 & A8 & A9 & A10 --> SA
    SA --> CG
    CG -- "Medium/Low" --> R1 & R2
    CG -- "High" --> R3
    R1 & R2 --> R3
    R3 --> D4
    D4 --> Final[Final Trading Plan]

    style A1 fill:#9B8AAB,stroke:#8B7A9B,color:#fff
    style A2 fill:#9B8AAB,stroke:#8B7A9B,color:#fff
    style A3 fill:#9B8AAB,stroke:#8B7A9B,color:#fff
    style A4 fill:#9B8AAB,stroke:#8B7A9B,color:#fff
    style A5 fill:#9B8AAB,stroke:#8B7A9B,color:#fff
    style A6 fill:#7A9A7A,stroke:#6B8E6B,color:#fff
    style A7 fill:#7A9A7A,stroke:#6B8E6B,color:#fff
    style A8 fill:#7A9A7A,stroke:#6B8E6B,color:#fff
    style A9 fill:#7A9A7A,stroke:#6B8E6B,color:#fff
    style A10 fill:#7A9A7A,stroke:#6B8E6B,color:#fff
    style SA fill:#B8A082,stroke:#A89072,color:#fff
    style CG fill:#7A9FB3,stroke:#6B8FA3,color:#fff
    style R1 fill:#7A9A7A,stroke:#6B8E6B,color:#fff
    style R2 fill:#C4A484,stroke:#B49474,color:#fff
    style R3 fill:#7A9FB3,stroke:#6B8FA3,color:#fff
    style D4 fill:#7A9A7A,stroke:#6B8E6B,color:#fff
    style Final fill:#C4A484,stroke:#B49474,color:#fff

System Capabilities: ✓ Multi-agent orchestration with LangGraph (parallel analyst execution) ✓ Confidence-gated research debate (HIGH skips debate; 30–70K tokens saved/session) ✓ Deterministic Risk Overlay — zero LLM cost position sizing and stop-loss ✓ Signal accuracy system — entropy confidence, VIX interpolation, adaptive weights ✓ Walk-forward backtesting with MAE/MFE per-trade analysis ✓ Background market scanner with opportunity detection and auto-analysis ✓ Real-time price feed via WebSocket ✓ Vector memory with pgvector ✓ Multiple LLM provider support (OpenAI, Anthropic, Gemini, Ollama, OpenRouter) ✓ Next.js 16 web application (sessions, screener, portfolio, watchlists, alerts) ✓ Production-ready Docker deployment ✓ Comprehensive observability stack (Prometheus + Grafana)