This repository offers a Python framework and progressive examples for building sophisticated AI agents. It targets developers and researchers seeking to create agents with multi-layered memory, task planning, sub-agent delegation, and team collaboration, leveraging the Anthropic Claude API. It provides a practical path from basic API interaction to complex agent orchestration.

How It Works

The core is a modular Python framework for AI agents. It features a three-layer memory system with automatic compression for efficient context management. Task execution uses a planning system with a todolist and a dispatch mechanism for delegating tasks to specialized sub-agents operating in isolated contexts, enabling parallel execution. Agent Team provides persistent, role-based collaboration via a file-based inbox. External tools integrate via the Model Context Protocol (MCP).

Quick Start & Requirements

Install dependencies via pip install -r requirements.txt after setting up a Python virtual environment. Configure ANTHROPIC_API_KEY in .env. Launch the main agent with python agent.py or tutorials with python build-agent-example/code/step01_single_call.py. A valid Anthropic API key is mandatory.

Highlighted Details

• Multi-Layered Memory: Working, contextual, long-term memory with automatic compression and startup archiving.
• Sub-Agent Dispatch: Delegates tasks to specialized sub-agents in isolated contexts, supporting concurrent execution.
• Agent Team: Enables persistent, role-based collaboration with independent threads and message queuing via .team/inbox/.
• MCP Integration: Connects to and utilizes external tool servers through the Model Context Protocol.
• Pluggable Skills: Dynamically loads skills defined in SKILL.md files.
• Task Planning: Integrates a todolist mechanism (update_todos tool) for structured task management.

Maintenance & Community

Specific details regarding maintainers, community channels (e.g., Discord, Slack), or active sponsorships are not explicitly detailed in the provided README.

Licensing & Compatibility

The README does not specify a software license, requiring further investigation for usage rights, especially for commercial applications.

Limitations & Caveats

Sub-agents cannot dispatch other sub-agents or directly modify the main agent's task list. Concurrency is model-scheduled based on tool call patterns, potentially requiring explicit instructions for parallel execution. The system relies on the Anthropic API, incurring costs and provider dependency. The absence of a stated license is a significant adoption blocker.