GAWD: A Graph-Model-Based Visual Agent Workflow Design Framework

GAWD (Graph-based Agentic Workflow Design) is a no-code web platform that helps users design and execute agent workflows through visual graph-model-based methods, supporting exports to CrewAI and PydanticAI frameworks. Its core value lies in lowering the barrier to agent workflow development, enabling non-technical users to quickly build complex AI autonomous systems while maintaining flexibility to adapt to multi-scenario needs.

With the improvement of Large Language Model (LLM) capabilities, AI agents are evolving towards complex autonomous systems. However, designing reliable and scalable workflows faces challenges such as multi-agent collaboration, state management, conditional routing, etc. Traditional coding methods are flexible but have high barriers, making them difficult for non-technical users or rapid prototyping scenarios. The GAWD framework thus emerged to provide a visual graph model solution.

GAWD abstracts workflows into directed graphs, where nodes represent operation units and edges represent processes. Core node types include:

• Start Node: Unique entry point, initializes global state variables;
• Agent Node: Core node, configures LLM, prompts, memory, tool calls, etc.;
• Routing Node: Implements conditional branching logic;
• Action Node: Modifies global state variables;
• End Node: Marks the end point, defines the final output. In addition, global state variables support three types: numeric, string, and boolean, and nodes can access input variables (predecessor outputs).

GAWD has complete execution capabilities:

1. Model Validation: Checks graph correctness before execution (no unconnected nodes, all paths lead to end nodes, type consistency, etc.);
2. Multi-Framework Support: Automatically generates Python code for CrewAI (multi-agent collaboration) and PydanticAI (strong type validation);
3. Execution Terminal: Real-time monitoring of progress, input values, viewing outputs and decision processes, facilitating debugging.

GAWD is suitable for the following scenarios:

• Rapid Prototyping: Non-technical personnel (product managers, analysts) can validate business logic without code;
• Complex Decision-Making Processes: Visual design makes the logic of multi-step, conditional branching, and manual review clear;
• Multi-Agent Collaboration: Build layered systems through agent handovers;
• Educational Demonstration: Help students understand the principles of agent workflows.

GAWD's frontend is built with React, providing a drag-and-drop editing experience; the backend supports user registration and session persistence, allowing workflows to be saved and loaded; the interface uses the Heroicons icon library to maintain a beautiful and consistent look.

GAWD transforms complex agent workflows into intuitive visual operations through graph model abstraction, lowering development barriers while supporting complex scenarios. As agent technology develops, such low-code/no-code platforms will play an important role in the popularization of AI applications, helping more teams explore the potential of agents.