I built a chaos engineering platform for AI agents. Here's why.

I've spent years building systems that couldn't afford to fail. Payments infrastructure processing millions of transactions a month teaches you one thing fast: reliability isn't optional, it's the product.

When I moved into AI/ML engineering, I kept waiting for chaos engineering tooling built for AI agents. It never showed up.

Teams are shipping LangChain pipelines, CrewAI workflows, OpenAI Agents SDK integrations, and MCP-powered systems into production every day. But nobody has a good answer to what happens when something breaks mid-run.

So I built Sepurux.

Fault injection, adversarial simulation, and stress testing for AI agent workflows in about 30 seconds. No ripping apart your existing stack. Works with LangChain, LangGraph, CrewAI, OpenAI Agents SDK, LlamaIndex, Pydantic AI, and MCP workflows. SDKs in Python, TypeScript, and Go.

Building this in public. Early days, but the core is working. Looking for teams shipping agents into production who want to try it and give feedback.

If that's you, drop a comment or reach out.