HYDRA

Multi-agent orchestration system for autonomous software development

╦ ╦╦ ╦╔╦╗╦═╗╔═╗
╠═╣╚╦╝ ║║╠╦╝╠═╣
╩ ╩ ╩ ═╩╝╩╚═╩ ╩

---

What is Hydra?

Hydra orchestrates multiple AI agents to build production software in parallel. Give it a project description, and it creates tickets that agents work on simultaneously - respecting dependencies, preventing conflicts, and ensuring quality.

Quick Start

Installation

Requirements: Python 3.12+, tmux, Redis (for distributed features)

Option 1: pipx (Recommended)

# Install pipx if not already installed
# Ubuntu/Debian: sudo apt install pipx
# macOS: brew install pipx
# Or: python3 -m pip install --user pipx

# Ensure pipx is in PATH
pipx ensurepath
# Restart terminal or: source ~/.bashrc

# Install Hydra
git clone https://github.com/privkeyio/hydra.git
cd hydra
pipx install .

# For development
pipx install -e .

Option 2: Virtual Environment

git clone https://github.com/privkeyio/hydra.git
cd hydra
python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -e .

Option 3: Direct pip install

git clone https://github.com/privkeyio/hydra.git
cd hydra
pip install -e .

Basic Setup

# Choose your LLM provider (claude_tmux, venice, mock)
export LLM_PROVIDER=venice  # or claude_tmux

# For Claude provider
export CLAUDE_CLI_PATH=/path/to/claude
# OR
export ANTHROPIC_API_KEY=your_key

# For Venice provider (recommended for open-source models)
export VENICE_API_KEY=your_key  # Get from https://venice.ai

Your First Project

# 1. Generate tickets from your idea
hydra ticket create "Build a REST API with authentication, database, and tests" --output tickets.yaml

# 2. Start the dashboard (auto-detects project)
hydra dashboard start

# 3. Execute tickets (dashboard auto-updates)
hydra ticket execute tickets.yaml 001

# 4. View real-time progress at http://localhost:8080

Core Commands

Single Task

hydra claude execute "Create a FastAPI service with JWT auth"

Ticket Workflow

# Generate tickets
hydra ticket create "project description" --output tickets.yaml

# Execute single ticket
hydra ticket execute tickets.yaml 001

# Execute all tickets in parallel
hydra ticket parallel tickets.yaml --workers 4

# Verify parallel execution results
hydra ticket verify-parallel tickets.yaml --workers 2

Session Management

hydra claude session /project --name dev
hydra claude list
hydra claude attach dev

How It Works

1. Create Tickets: Break down your project into atomic tasks with dependencies
2. Parallel Execution: Multiple agents work simultaneously on different tickets
3. Smart Coordination: Dependency resolution, file locking, conflict prevention
4. Quality Gates: Automatic testing, linting, and verification
5. Live Monitoring: Real-time dashboard shows progress and logs

Example Ticket

tickets:
  - id: "001"
    title: Setup API
    status: TODO
    priority: 1
    model: balanced
    description: Create FastAPI application with core structure
    acceptance_criteria:
      - Create main.py with FastAPI app
      - Add health check endpoint at /health
      - Setup global error handling middleware
      - Add CORS configuration
    dependencies: []

Key Features

🚀 Performance Optimizations

• uvloop integration for 2-5x async performance boost
• Connection pooling with singleton patterns for Redis/HTTP
• Smart batching for parallel LLM requests
• Response caching with configurable TTL

🛡️ Safety & Quality

• AI detection to flag generated code patterns
• Quality gates with automatic linting and testing
• Safety guards blocking dangerous operations
• Graceful shutdown with proper cleanup sequence

📊 Dashboard & Monitoring

Quick Start:

hydra dashboard start  # Auto-detects project, runs at http://localhost:8080
hydra dashboard stop   # Stop the dashboard

Features:

• Auto-detects project - Uses project database when run from project directory
• Real-time updates - Status changes reflected within 5 seconds
• Persistent - Runs independently, survives hydra restarts
• Token tracking - Budget management and cost warnings
• Cost reporting with detailed breakdowns
• Database backend for persistence and queries

🔧 Enhanced Workflow

• Smart ticket creation with codebase analysis
• Context sharing between dependent tickets
• .hydra directory for organized project data
• Modular CLI with improved help and documentation

Provider Support

• Claude (claude_tmux) - Interactive Claude Code CLI via tmux
• Venice AI (venice) - Production-ready with retry logic and streaming
• Anthropic (anthropic) - Direct API integration
• OpenAI (openai) - GPT-4 and GPT-3.5 support
• Mock (mock) - Testing and development provider

Documentation

• 📚 Full Documentation - Complete reference and guides
• 🏗️ Architecture - System design and components
• 🔧 API Reference - Python API and integration
• 💡 Examples - Real-world use cases

Testing Commands

Try these to see Hydra in action:

# Build a complete web app
hydra ticket create "Build a todo app with React frontend and FastAPI backend" --output tickets.yaml
hydra ticket parallel tickets.yaml --workers 4

# Run quality checks on implementation
hydra quality 001 --strict

# Verify all work is complete
hydra ticket verify-parallel tickets.yaml --workers 2

# Check the dashboard
open http://localhost:8080

# Check the results
ls -la .hydra/reports/
cat tickets.yaml

License

This project is licensed under the MIT License. See LICENSE for details.

---

Built for production. No compromises.

HYDRA - Orchestrating the future of autonomous development.