Maestro is a Python framework designed to orchestrate multiple AI agents, primarily for task decomposition and execution. It targets users who need to automate complex workflows by breaking down objectives into smaller, manageable sub-tasks, executing them, and then refining the results. The primary benefit is enabling AI models, particularly Anthropic's Claude Opus and Haiku, to intelligently manage and execute multi-step processes, with added support for other LLMs via LiteLLM.

How It Works

Maestro employs a multi-stage AI workflow. An "orchestrator" model (e.g., Claude Opus) breaks down a user-defined objective into discrete sub-tasks. Each sub-task is then executed by a "sub-agent" model (e.g., Claude Haiku), which is provided with context from previously completed tasks. Finally, the orchestrator model reviews and refines the results from all sub-tasks into a cohesive final output. This iterative decomposition and execution approach allows for complex problem-solving by leveraging specialized AI roles.

Quick Start & Requirements

• Install dependencies: pip install litellm (or pip install -r requirements.txt if cloning the repo).
• Set environment variables for API keys (e.g., OPENAI_API_KEY, ANTHROPIC_API_KEY).
• Run with: python maestro.py (or python maestro-anyapi.py, python maestro-gpt4o.py, python maestro-lmstudio.py, python maestro-ollama.py, python maestro-groq.py for specific integrations).
• Prerequisites: Python, API keys for chosen LLM providers. Optional: Tavily API key for search functionality.
• Local LLM support via LMStudio or Ollama requires separate setup and model downloads (e.g., ollama.pull('llama3:70b')).

Highlighted Details

• Supports Anthropic Claude 3.5 Sonnet, Opus, Haiku, and various models via LiteLLM (OpenAI, Gemini, Cohere, Groq).
• Integrates optional web search (Tavily) for enhanced sub-task information gathering.
• Capable of generating code files and folders for coding-related projects.
• Logs the entire interaction, including task breakdown and execution, to a Markdown file.

Maintenance & Community

The project appears actively maintained, with recent updates for Claude 3.5 Sonnet and GPT-4o support. Links to community resources like Discord or Slack are not explicitly provided in the README.

Licensing & Compatibility

Released under the MIT License, which permits commercial use and linking with closed-source projects.

Limitations & Caveats

The effectiveness of the task breakdown and execution is highly dependent on the quality of the prompts and the capabilities of the chosen LLMs. API costs can accumulate quickly with complex or lengthy objectives. Search functionality requires a Tavily API key, which is a separate service.