Show HN: Grapheteria - A structured workflow framework for agent orchestration
github.comI know what you're thinking, "Oh no, not ANOTHER agentic workflow library ". I felt the same way, but hear me out on why we've finally hit the sweet spot.
We've all been caught between 2 frustrating options:
- Code-only frameworks: Powerful but often buried under layers of abstractions
- UI-only builders: Great for simple flows but hit a wall when you need real customization
Code is non-negotiable. Visual debugging is invaluable. I built Grapheteria to bridge the gap.
Edit your code, visualize the workflow instantly Interact with your flow visually, code updates right away.
Grapheteria follows a simple principle: design clean, composable graphs where each node and edge has a clear purpose.
Out-of-the-box features include human-in-the-loop, step-by-step debugging, and solid logging. Everything else is just an API call away.
Here's how Grapheteria makes your life easier as an AI developer:
- Zero abstraction tax - the code you write is the code that runs
- Visually debug in the UI by time traveling. Made a mistake? Simply step backwards, fix it and step forward again! No restarts necessary.
- Grapheteria is natively wrapped in a webserver. Place it in a container, host it anywhere and send events to your workflow with simple HTTP requests!
- Integrates seamlessly with ecosystem innovations like MCP and A2A
Check it out and feel free to show some love : https://github.com/beubax/Grapheteria
What are your thoughts on graph-based workflow systems? And what's been your experience with the code vs. UI tradeoff in other tools?
How does Grapheteria manage synchronization between visual workflow design and code, ensuring that updates in the UI are accurately reflected in the underlying codebase, and vice versa?
Websockets and observers. An observer monitors the current working directory, and when any file changes, it triggers an event. The websocket then broadcasts the updated code to all connected UIs. Any action performed in the UI modifies the codebase directly, never the UI state itself. These file changes trigger the observer, completing the loop. This architecture ensures the codebase remains the single source of truth, with UI changes always reflecting the actual code state rather than optimistic predictions.