AgentFlow: A Visual AI Agent Workflow Orchestration Platform

AgentFlow is an open-source AI agent workflow orchestration platform that allows users to create AI agents, connect MCP tools, and design automated workflows via a visual editor. Built on React, NestJS, and PostgreSQL, it supports various trigger and node types, providing a complete engineering solution for AI automation. It addresses engineering challenges such as AI agent collaboration management, visual workflow design, and external tool integration, enabling non-technical users to participate in automated workflow design.

With the rapid development of Large Language Model (LLM) capabilities, AI agents have become important tools for automated task execution. However, scaling them into complex workflows faces many challenges: How to manage collaboration among multiple agents? How to visually design execution flows? How to integrate external tools and data sources? Traditionally, developers need to write a lot of code to orchestrate agent interactions, increasing development costs and making it difficult for non-technical users to participate. AgentFlow was created to solve these problems, providing a production-grade visual platform that makes building AI workflows as simple as building blocks.

Technical Architecture

The project uses a Monorepo architecture with the following tech stack:

Layer	Tech Stack
Frontend	React 18 + Vite + TypeScript + TailwindCSS + ShadCN UI
State Management	React Query + Zustand
Backend	NestJS + TypeScript
Database	PostgreSQL + Prisma ORM
Project Management	Turborepo + pnpm

Agent Management

Users can create and manage multiple AI agents with configurations including:

• Basic settings: Name, description, system prompt
• Model parameters: Temperature value, model selection
• Tool integration: Enable/disable tools, configure MCP server

MCP (Model Context Protocol) Management

MCP is a core feature that defines a standardized protocol for communication between agents and external tools. It supports:

• Predefined MCP servers
• Dynamic addition of third-party services
• Security authentication configuration (API key, token) Through MCP, agents can call external APIs, query databases, and expand their capability boundaries.

The platform provides a drag-and-drop visual workflow builder with rich node types: Trigger nodes: Manual trigger, scheduled trigger (Cron expression), Webhook trigger Control flow nodes: Conditional branch, multi-branch, merge node AI nodes: Execute agent, agent decision, agent review Tool nodes: Execute MCP tool, get MCP resource Utility nodes: Delay, set variable, data transformation, JSON parser, response, notification, log The node-based design allows non-technical users to participate in workflow design while meeting the needs of complex scenarios.

AgentFlow is suitable for various automation scenarios:

• Content creation pipeline: Scheduled trigger to generate article drafts, auto-publish to multiple platforms after manual review
• Customer service automation: Webhook receives inquiries, agent provides initial reply, transfers complex issues to humans, sends satisfaction survey after completion
• Data analysis report: Scheduled pull from multiple data sources, agent analyzes to generate insights, converts to PDF and sends via email
• DevOps automation: Receives GitHub Webhook, agent analyzes code changes, triggers tests, decides whether to deploy to production

Technical Implementation Highlights

• Type safety: Full project developed with TypeScript, complete type definitions from database to API interface, Prisma ensures safe database operations
• Performance optimization: Vite accelerates development, React Query cache reduces requests, Turborepo remote cache speeds up CI/CD
• Extensibility: Modular architecture facilitates adding new nodes, MCP standard interface simplifies third-party integration

Project Roadmap

Planned features include:

• Workflow template market
• Workflow import/export
• MCP application market
• Multi-person collaborative editing
• AI-generated workflows These plans demonstrate the project team's deep understanding of the AI automation field.

AgentFlow represents an important direction in AI agent engineering: abstracting complex orchestration logic into visual components, lowering the usage threshold while maintaining professional flexibility. For teams looking to integrate AI into business processes, it is an open-source solution worth evaluating. As AI capabilities evolve, such platforms will play a more important role in enterprise automation transformation; open-source features promote community participation and drive the democratization of AI automation technology.