Hey everyone!
I’ve been building StepFork, an open-source, local-first tool for behavioral regression testing of AI agents.
The idea is simple: when your agent fails, turn that failure into a reproducible test.
StepFork captures agent execution traces in .sftrace files, lets you replay and evaluate them, and helps identify behavioral regressions as you change your agent or its underlying models.
I’m looking for people working on LLM applications, AI agents, RAG systems, or agentic workflows to try it out and share honest feedback.
I’d especially appreciate feedback on:
Getting started: Is the installation and workflow straightforward?
Usability: Does the CLI feel intuitive? What’s confusing or missing?
Real-world usefulness: Would this help you catch regressions in your own agents?
Bugs and edge cases: What breaks when you try it with your own workflows?
Feature direction: What would make you actually use a tool like this?
You don’t need to contribute code. Even a quick review, a small experiment, or a GitHub issue describing your experience would be incredibly helpful.
I’m building this in the open and would rather get honest criticism now than build features nobody needs.
If you’re interested, give it a try and let me know what you think! Thanks for your time and feedback.
#AI #LLM #Agents #OpenSource #MachineLearning