agentic-ai-systems

Agentic AI Systems is an orchestration framework that models an agent hierarchy where a Main Agent can spawn Subagents to handle tasks in parallel or specialized ways. The project emphasizes dynamic agent orchestration, where the Main Agent delegates subtasks to Subagents (like a hen with her chicks) and aggregates results to produce a final response. Use this MCP server to run the main orchestrator, which can leverage predefined workflows (e.g., Baseline, Prompt Chaining, Routing, Parallelization, Orchestrator-Workers, Evaluator) as described in the documentation, or to explore Autonomous Agent patterns for open-ended problems. The server exposes capabilities to spawn subagents, route tasks, parallelize work, and evaluate results, enabling complex multi-step problem solving and task decomposition. To operate, start the server in the environment where the necessary runtime is installed, then interact with the controller to submit tasks and receive orchestrated outputs. Tools and capabilities include: spawning subagents, coordinating multi-step workflows, routing classifications, compiling results from parallel tasks, and applying evaluator iterations to improve output quality.

Prerequisites:

• Python 3.11+ (or compatible Python environment)
• Access to required API keys (e.g., OpenAI or Claude if used by the agents)
• Git to clone the repository

Installation steps:

1. Clone the repository git clone https://github.com/thibautmelen-agentic-ai-systems.git cd thibautmelen-agentic-ai-systems

2. Set up a Python virtual environment python -m venv venv source venv/bin/activate # macOS/Linux venv\Scripts\activate # Windows

3. Install dependencies pip install -r requirements.txt

4. Configure environment variables

   • Create a .env file or export variables in your shell API_KEY=your-api-key-here OPENAI_API_KEY=your-openai-api-key LOG_LEVEL=INFO

5. Run the MCP server python -m agentic_ai_systems

Notes:

• If the project ships a different entry point, adjust the command in mcp_config accordingly.
• Ensure network access to any external APIs required by subagents.
• For development, use a virtual environment to isolate dependencies.

Tips and common issues:

• Ensure API keys are valid and have the necessary permissions for the LLMs/tools used by subagents.
• If subagents fail to spawn, verify that the main agent has permission to spawn child processes in the runtime environment.
• Adjust LOG_LEVEL to DEBUG during debugging to gather more diagnostic information.
• When using routing or parallelization workflows, monitor rate limits and concurrency settings to avoid throttling.
• Keep the environment variables in a secure place and avoid hard-coding sensitive keys in code or config files.
• If upgrading dependencies, test the orchestration flow with a small task before running large campaigns with multiple subagents.