RunVeto – A Simple Kill Switch for Autonomous AI Agents

Hey HN,

I’m a software engineer with 5 years of experience, and I’ve been building autonomous agents recently using LangChain. I noticed that we are giving these agents too much "autonomy" without enough "governance".

The specific problem I faced was Agent Sprawl/Recursion. A test bot got stuck in a recursive loop and almost ran up a significant OpenAI bill before I noticed and killed the process manually.

So, I’m building RunVeto.xyz as a minimal governance layer—a control plane for agent guardrails. It sits between your agent and the LLM API, framework-agnostic, and integrates with one line of code.

What I’m planning to implement:

Hard-Cap Budgeting: Set strict token/cost limits to kill any task before it breaks the bank.

Global 'Veto' Button: Pause or terminate any active agent process from a central dashboard.

PII Shield: Automatic scrubbing of sensitive data (PII) before it hits the LLM.

'Chain-of-Thought' Audit: Real-time visibility into agent planning logs.

I'm currently pre-MVP and using this landing page to validate the core features and find "founding developers" to guide the roadmap.

I'd love to hear this community's critique. Have you encountered "recursive loops" in your own agent workflows? What’s your biggest operational nightmare with agents?

The landing page has an embedded survey. I’m eager to hear your thoughts.

No fluff, just safety.