RunawayGuard

Real-time circuit breaker for runaway AI agent costs

Problem

The biggest fear when building agents is a loop bug or hallucination blowing up your API bill. A recent Show HN ('AtlasBurn – know what your AI costs and stop runaway agents') confirms this is a real and urgent pain. Existing tools either only report post-hoc, or cover a single provider, with no cross-model real-time cutoff.

Solution

A lightweight proxy (local or self-hosted) that sits in front of OpenAI/Anthropic/Gemini APIs, accumulates token costs in real time, and hard-blocks requests when a budget threshold is hit. Set budgets per project or per agent session, with a simple dashboard and history.

Users

Indie developers and small teams building LLM agents, plus researchers running automated experiments

MVP

Local HTTP proxy that intercepts major AI APIs; per-session/project cost tracking; auto 429 + alerts on budget breach; minimal web dashboard; one-click Docker deploy

Revenue

Open-source core with hosted cloud subscription; team plan priced by number of agents or monthly calls

Why now

Agent apps are exploding and runaway-cost stories are common. Big clouds won't build a 'breaker' that reduces their own revenue — a perfect gap for an indie hacker.