Agentsflow provides a drag-and-drop web interface for building and executing autonomous AI agent workflows using Microsoft's AutoGen framework. It aims to simplify agent creation and interaction for beginners while offering advanced customization for experienced users, enabling visual flow design and runtime interaction with agents.

How It Works

The project leverages a Next.js frontend with React Flow for the visual interface, allowing users to connect and manage AutoGen agents. It utilizes WebSockets for real-time communication between the frontend and the backend agent execution environment, facilitating dynamic agent control and interaction during runtime.

Quick Start & Requirements

• Install using git clone https://github.com/jaemil/agentsflow, cd agentsflow, npm install, and npm run agentsflow.
• Requires Git, Node.js, Python 3.11, and Poetry.
• Access the UI at localhost:4200.
• Development server: npm run agentsflow:dev.
• Official quick-start: https://github.com/jaemil/agentsflow

Highlighted Details

• Visual drag-and-drop interface for agent flow creation.
• Real-time interaction and monitoring of running agents.
• Supports agent code execution.
• Next.js frontend with React Flow integration.

Maintenance & Community

• Active development, currently in initial stages.
• Discord server available for discussion and contributions.
• GitHub issues for tracking requests and bugs.

Licensing & Compatibility

• Licensed under GNU General Public License (Version 3).
• GPLv3 is a strong copyleft license, potentially restricting integration with closed-source commercial applications.

Limitations & Caveats

The project is in its initial stages with many core features still marked as "to-do" on the roadmap, including vector database integration, global settings, agent customization, and saving/exporting flows.