Skip to main content

Visual Agent Builder

Experimental & Educational Tool

This Visual Agent Builder is an experimental feature provided primarily for educational purposes to support our course curriculum. It is subject to rapid changes and is not guaranteed to be bug-free. Use it to learn and prototype ADK architectures.

📺 Watch the Demo​

Check out the full walkthrough of the builder in action:


Design Visually. Deploy as Code.

The Visual Agent Builder is a tool designed to bridge the gap between architectural diagrams and deployable code. It allows you to design complex Multi-Agent Systems using the Google Agent Development Kit (ADK) schema.

While the builder abstracts away the structural YAML syntax, certain advanced configurations or logic definitions might still require inputting specific code snippets or detailed settings as permitted by the AgentConfig feature set.

Purpose & Philosophy​

Building agents often starts on a whiteboard. You draw boxes for "Research Agent" and "Writer Agent," and draw arrows connecting them. Translating that whiteboard into the specific agent.yaml configuration required by the ADK can be tedious and error-prone.

Our Goal:

  1. Visual Abstraction: Drag, drop, and connect agents on an infinite canvas.
  2. Schema Compliance: The builder strictly adheres to the official Google ADK AgentConfig schema. It ensures your configuration is valid before you even export.
  3. Modular Architecture: The builder automatically generates a clean, multi-file project structure, separating concerns between the root orchestrator and individual agents.

Core Features​

1. The Canvas​

An infinite, zoomable workspace where you can orchestrate complex flows. Whether you are building a simple chatbot or a cyclic research swarm, the canvas adapts to your complexity.

2. Supported Agent Types​

Currently, the builder supports the foundational unit of the ADK:

  • Llm Agent: The fundamental unit of work, powered by Gemini models.

Work in Progress: We are actively working on adding visual support for orchestrator agents:

  • Sequential Chain (Coming Soon)
  • Parallel Group (Coming Soon)
  • Loop Controller (Coming Soon)

3. Deployment​

The builder focuses on Configuration Generation. Once you have designed your agent:

  1. Click Export to push the configuration to GitHub or download a ZIP.
  2. Split-File Output: The export provides a full project folder.
    • agents/root_agent.yaml: The main orchestrator file.
    • agents/my_agent.yaml: Individual agent definitions.
  3. Use the generated project in your local environment.
  4. Run the agent using the standard ADK CLI (adk run agents/root_agent.yaml) or deploy it to Google Cloud Run following standard ADK deployment procedures.

We will be adding specific "Run & Deploy" guides directly within the tool documentation as the feature set matures.

Documentation Series​

This documentation is structured to accompany our YouTube Series. Each section below corresponds to a video tutorial walking you through a specific feature.

  1. Building an Llm Agent - The "Hello World" of visual agents.
  2. (Coming Soon) Sequential Workflows - Chaining agents together.
  3. (Coming Soon) Parallel Processing - Reducing latency with concurrency.
  4. (Coming Soon) Tools & Functions - Giving your agents capabilities like Search and Code Execution.
Get Started

Click on Llm Agent to create your first design.