Inspiration

LLM based Agentic Workflows are touted to be the highlight of the year. I've seen first hand the their potential; they could totally disrupt the industry! However, to make it easy for people to onboard into the topic, we need to offer them easy tools to experiment and feel the "wow" factor for themselves. This project aims to help people build what agentic workflows are in a simple way.

What it does

It's a visual GUI where users can build agentic workflows using simple drag and drop of blocks (ie. block based builder). They can create graphs with multiple agents talking to each other, or working in isolation, define personas for agents and assign tasks. Created agentic workflows are then exported into code, which can be executed. The app is packaged as VS Code extension, making it easy for developers to consume. The overall idea is to take a low-code approach to agentic workflow building.

How we built it

I am a backend developer through and through, with experience in DevOps. So one personal challenge for me was to build a decent UI :). I used Reach + Vite + React Flow with Typescript for the GUI, and packaged it as a VS Code extension.

The graph that you build with the editor gets complied into a crewAI project. crewAI is an awesome project to experiment with agents.

Challenges we ran into

Well, everything related to UI was pretty much a challenge for me. And figuring out the sweet spot between what I can do in 24 hours (actually lesser) vs what's ambitious.

Accomplishments that we're proud of

The fact that I was able to build a PoC as per my initial plan (and with a decent UI :) ) is something for me to be proud of. In fact I think that this project has a lot of potential, and I will be taking this forward with the help of the community. Looking forward to the feedback when I present it.

What we learned

Lots, but in two sentences:

  • UI is hard but so rewarding if done well. It makes the difference between hard and easy :)
  • Packaging the application is an interface closed to developers is the way to go!

What's next for AgentBuddy

It's already opensource. I'd like to keep building on it to support other frameworks like Langgraph as targets, as well as to visualize execution within the graph. Adding support for tools and agent hierarchies.

Built With

Share this project:

Updates