ANT-AI – Lightweight Framework for Multi-Agent AI Systems

Topics
Multi-Agent SystemsAI AgentsPythonLLMWorkflow AutomationOpen Source
Team Cezar Sas , Vincenzo Giuffrida , Sandra Mitrović , Matteo Salani
Links Website · Code

Overview

ANT-AI is a Python framework for building multi-agent AI systems — from a single tool-driven agent to a coordinated colony of collaborating peers — designed to be easy to pick up whether you are running research experiments or integrating autonomous agents into a real application.

Built around a graph-based workflow engine and native A2A (Agent-to-Agent) communication, agents delegate tasks to one another without any custom glue code. The LLM-agnostic core means you can swap models or providers at any point without touching agent logic, keeping experiments reproducible and production deployments flexible.

ANT-AI exposes exactly what runs, when, and why — giving researchers the transparency they need to publish and practitioners the control they need to ship.

Key capabilities include:

  • Graph-based workflow orchestration for structured, auditable agent pipelines
  • Native A2A communication for agent-to-agent delegation and multi-agent collaboration
  • MCP tool integration with a built-in registry and support for custom tools
  • Memory backends including Mem0 integration for persistent agent state across interactions
  • Built-in observability via Langfuse tracing and lifecycle hooks for guardrails and monitoring

ANT-AI is available on PyPI (requires Python 3.14+):

uv add ant-ai