Hey Product Hunt! 👋

I'm Pavel, the creator of AGBoss.

I've spent over 10 years building enterprise applications in Java, working extensively with Spring and dependency injection.

When I started building AI agents, something felt familiar: the ecosystem reminded me of enterprise Java before frameworks like Spring made application architecture more structured and manageable.

But there was another problem that bothered me even more.

A TypeScript agent workflow can compile successfully and still be structurally broken.

A required piece of data might not be available on a particular execution path. The types look fine, but the workflow fails when that path is actually executed.

That's why I built AGBoss.

It's an open-source TypeScript framework built on LangGraph, with:

Static workflow verification — agboss check detects structural problems before execution.
Declarative composition — describe workflows and their connections explicitly.
Spring-style dependency injection — structure agent applications using familiar architectural patterns.
Isolated deterministic testing — test components without running an entire multi-agent workflow.

My goal isn't to create yet another abstraction around LLM calls. I want complex agent systems to become easier to reason about, validate, test, and maintain.

AGBoss is still early (v0.1.0), and I'd particularly appreciate feedback from engineers building real agent workflows.

A few questions I'd love your thoughts on:

What structural bugs have you encountered in multi-agent workflows?
How are you currently testing individual agents and workflow branches?
Would static validation change how you develop AI applications?

GitHub: https://github.com/agbossjs/agboss

Docs: https://agboss.dev/

Thanks for taking a look! I'm happy to discuss the architecture, implementation details, or trade-offs.