Docs: https://ashp15205.github.io/guardian-runtime/

Hi Guys,

I built Guardian Runtime: a local FinOps and security proxy for AI coding agents (like Claude Code, Cursor, and Aider).

You point your agent to localhost:8080 instead of the real LLM API, and it sits in the middle to do three things:

1. Hard Budgets: Auto-kills the agent's internet access if it gets stuck in a retry loop and hits your daily spend limit. 2. Local Scanners: Instantly drops outgoing requests that contain leaked API keys or PII. 3. Terse Mode: Optimizes system prompts to force shorter answers, cutting your API output costs by 40–70%.

It runs entirely on your machine with zero cloud dependency, and supports OpenAI, Anthropic, and Gemini out of the box.

pip install guardian-runtime

I'd love to hear your feedback or see if you can break it!