Multi-Agent Coding Assistant

A sophisticated multi-agent coding assistant powered by LangGraph and Ollama, designed to help with complex software development tasks through specialized AI agents.

🚀 Features

Multi-Agent Architecture: Coordinated agents working together for complex tasks
Local AI Models: Uses Ollama for privacy and control over your AI models
Per-Agent Model Optimization: Different models optimized for each agent's specialty
Automatic Model Management: Download required models on first run
Automatic Background Indexing: Real-time codebase analysis with file watching
Flexible Indexing Control: Enable/disable automatic indexing with CLI options
Command Safety Controls: Three security modes for system command execution
Specialized Agents (10 total):
- 🔍 WebSearchAgent (llama3.1:8b): Searches for documentation, examples, best practices
- 🔍 CodeReviewAgent (qwen2.5-coder:32b): Analyzes code quality, bugs, security issues
- 📊 CodeAnalyzerAgent (qwen2.5-coder:32b): Indexes codebases, tracks dependencies, AST analysis
- 📁 FileOperationsAgent (qwen2.5-coder:32b): Direct file manipulation - read, write, edit files
- 🧪 TestGeneratorAgent (qwen2.5-coder:32b): Creates comprehensive unit and integration tests
- 🔄 RefactoringAgent (qwen2.5-coder:32b): Code optimization, design patterns, modernization
- 📝 GitAgent (llama3.1:8b): Git workflows, commit messages, merge conflict resolution
- 📚 DocumentationAgent (llama3.1:8b): README files, API docs, docstrings, user guides
- 🖥️ CommandLineAgent (llama3.1:8b): System command execution with safety controls
- 🎯 SupervisorAgent (qwen2.5:32b): Orchestrates multi-agent workflows using LangGraph

📋 Prerequisites

Python 3.9+
Ollama installed and running locally
qwen2.5-coder:32b model (or your preferred coding model)

🛠️ Installation

Clone the repository:

git clone git@github.com:adam-hanna/ollama-agentic-coder.git
cd ollama-agentic-coder

Create a virtual environment:

python -m venv .venv
source ./.venv/bin/activate  # On Windows: .venv\Scripts\activate

Install dependencies:

pip install -r requirements.txt -r requirements-dev.txt

Install Ollama (if not already installed):

# macOS
brew install ollama

# Linux
curl -fsSL https://ollama.ai/install.sh | sh

# Windows - Download from https://ollama.ai/download

Pull the models:

ollama pull qwen2.5-coder:32b
ollama pull qwen2.5:32b
ollama pull llama3.1:8b

⚙️ Configuration

The system can be configured via environment variables or the default config:

Environment Variables

# Ollama settings
export OLLAMA_HOST="http://localhost:11434"
export OLLAMA_MODEL="qwen2.5-coder:32b"
export OLLAMA_TEMPERATURE="0.1"
export OLLAMA_MAX_TOKENS="4096"
export OLLAMA_TIMEOUT="300"

Available Models

You can use any Ollama-compatible model:

# Other excellent coding models
ollama pull deepseek-coder:33b
ollama pull codellama:34b
ollama pull starcoder2:15b

Command Line Safety Modes

The CommandLineAgent operates in three safety modes:

SAFE (default): Only pre-approved read-only commands (ls, cat, grep, etc.)
WHITELIST: Safe commands + user-approved commands (prompts for approval)
YOLO: All commands allowed (use with extreme caution)

Configure via environment variable:

export COMMAND_SAFETY_MODE="safe"  # or "whitelist" or "yolo"

🚀 Usage

Basic Usage

Run ollama:

ollama serve

Run the application:

python main.py

CLI Options

python main.py --help
python main.py --host http://localhost:11434 --model qwen2.5-coder:32b --temperature 0.1

# Indexing control options
python main.py --no-auto-index        # Disable automatic background indexing
python main.py --no-file-watch        # Disable file change watching
python main.py --no-auto-index --no-file-watch  # Disable both

Interactive Commands

Once running, you can use these commands:

/help - Show available commands and indexing status
/index - Manual indexing commands:
- /index - Index current directory
- /index [path] - Index specific directory
- /index auto - Toggle automatic background indexing
- /index stop - Stop background indexing
/config - Show current configuration
/agents - List available agents and their assigned models
/clear - Clear conversation history
/exit - Exit the application

Example Interactions

Code Review:

You: Review this Python file for bugs and security issues: src/auth.py

File Operations:

You: Create a new config.py file with database settings
You: Edit the main.py file to add logging configuration

Test Generation:

You: Generate comprehensive unit tests for the User class
You: Create integration tests for the authentication workflow

Refactoring:

You: Analyze auth.py for refactoring opportunities and suggest improvements
You: Refactor this complex function to use better design patterns

Git Operations:

You: Generate a commit message for my current changes
You: Help me resolve this merge conflict in src/models.py

Documentation:

You: Create a README for this project
You: Add comprehensive docstrings to all functions in this file

Web Search:

You: Search for best practices for async Python error handling

Command Line Operations:

You: List all Python files in the current directory
You: Check the status of the development server
You: Run the test suite and show me the results

Complex Multi-Agent Tasks:

You: I need to implement user authentication. Create the models, generate tests, add documentation, and suggest a Git workflow.
You: Refactor my authentication system, generate tests for the changes, and update the documentation.

🏗️ Architecture

Agent Structure

SupervisorAgent (qwen2.5:32b) - LangGraph Orchestration
├── WebSearchAgent (llama3.1:8b) - DuckDuckGo + Web Search
├── CodeReviewAgent (qwen2.5-coder:32b) - AST + Static Analysis
├── CodeAnalyzerAgent (qwen2.5-coder:32b) - Tree-sitter + Indexing
├── FileOperationsAgent (qwen2.5-coder:32b) - File System Operations
├── TestGeneratorAgent (qwen2.5-coder:32b) - Test Creation & Coverage
├── RefactoringAgent (qwen2.5-coder:32b) - Code Optimization
├── GitAgent (llama3.1:8b) - Version Control Operations
├── DocumentationAgent (llama3.1:8b) - Documentation Generation
└── CommandLineAgent (llama3.1:8b) - System Command Execution

Data Flow

User Input → SupervisorAgent analyzes request
Task Analysis → AI determines which agent(s) are needed
Agent Selection → Route to appropriate specialized agent(s)
Background Processing → Real-time codebase indexing if enabled
Agent Execution → Specialized agents work on their tasks
Result Synthesis → SupervisorAgent combines results into coherent response

Model Optimization Strategy

Coding Tasks: qwen2.5-coder:32b - Specialized for code analysis, generation, review
General Tasks: llama3.1:8b - Lighter, faster for search, git, documentation
Orchestration: qwen2.5:32b - Strong reasoning for task coordination

🔧 Customization

Adding New Agents

Create a new agent class inheriting from BaseAgent:

from core.base_agent import BaseAgent, AgentState

class MyCustomAgent(BaseAgent):
    def _default_system_prompt(self) -> str:
        return "Your specialized agent prompt here"
    
    async def process(self, state: AgentState) -> AgentState:
        # Your agent logic here
        pass

Custom Models

The system supports any Ollama model. You can customize per-agent models in core/config.py:

models: Dict[str, str] = {
    "supervisor": "qwen2.5:32b",
    "websearch": "llama3.1:8b", 
    "code_review": "qwen2.5-coder:32b",
    "file_operations": "qwen2.5-coder:32b",
    # ... customize as needed
}

Popular model choices:

qwen2.5-coder:32b - Excellent for coding tasks
deepseek-coder:33b - Strong coding capabilities
codellama:34b - Meta's coding model
starcoder2:15b - Smaller but capable
llama3.1:8b - Fast general purpose model
qwen2.5:32b - Strong reasoning for coordination

🧪 Testing

# Run the test suite
python -m pytest tests/

# Test specific components
python -m pytest tests/test_agents.py

📁 Project Structure

langgraph-ollama-agent/
├── core/                   # Core components
│   ├── config.py          # Configuration management
│   ├── ollama_client.py   # Ollama API client
│   └── base_agent.py      # Base agent class
├── agents/                # Specialized agents
│   ├── websearch_agent.py
│   ├── code_review_agent.py
│   ├── code_analyzer_agent.py
│   ├── file_operations_agent.py
│   ├── test_generator_agent.py
│   ├── refactoring_agent.py
│   ├── git_agent.py
│   ├── documentation_agent.py
│   ├── command_line_agent.py
│   └── supervisor_agent.py
├── cli/                   # Command line interface
│   └── main.py
├── utils/                 # Utility functions
├── tests/                 # Test files
├── requirements.txt       # Dependencies
├── main.py               # Entry point
└── README.md             # This file

🐛 Troubleshooting

Common Issues

Ollama Connection Error:

# Check if Ollama is running
ollama list

# Start Ollama if needed
ollama serve

Model Not Found:

# Pull the required model
ollama pull qwen2.5-coder:32b

Memory Issues:

Use smaller models like qwen2.5-coder:14b or starcoder2:7b
Reduce max_tokens in configuration

Slow Performance:

Use GPU acceleration if available
Reduce model size (use smaller variants like qwen2.5-coder:14b)
Disable background indexing: --no-auto-index
Disable file watching: --no-file-watch
Adjust temperature settings

🤝 Contributing

Fork the repository
Create a feature branch (git checkout -b feature/amazing-feature)
Commit your changes (git commit -m 'Add amazing feature')
Push to the branch (git push origin feature/amazing-feature)
Open a Pull Request

📄 License

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

🙏 Acknowledgments

LangGraph for the multi-agent orchestration
Ollama for local LLM serving
Qwen2.5-Coder for the excellent coding model
Tree-sitter for code parsing

adam-hanna/ollama-agentic-coder