How we built a local LLM traffic proxy to control token usage in our R&D workflow

We run Skilful Fox Studio, a self-funded R&D initiative focused on studying how intelligent systems adapt to engineering workflows. While building our core research track—an analytical AI profiling framework named Vulpis—we hit a massive, expensive roadblock: monitoring, caching, and controlling cloud LLM traffic.

Every developer tool, IDE extension, and internal script we tested wanted to talk directly to external APIs. There was no lightweight, local-first way to:

1. Log prompts across multiple disconnected tools in one centralized place.

2. Track exact token spending per environment or sub-task in real-time.

3. Cache repetitive LLM responses locally to save hard-earned R&D budget during heavy debugging cycles.

Enterprise API gateways were way too bloated and complex for a lean setup, so we decided to build a targeted utility for ourselves.

That’s how GreyFox was born. It’s a self-hosted, lightweight AI traffic proxy designed to sit directly on your local machine. It gives you absolute control, deep observability, and strict privacy over your LLM data streams without any telemetry or cloud dependencies.

We initially built it as an internal side-tool to support our own research, but today we launched the Community Edition because we believe other independent developers and small teams face the exact same token-bleeding and data-privacy issues.

The core functionalities are out, and it's completely free to use. We are currently analyzing market demand to see how other engineers adapt it to their environments, which will shape where we take this project toward the end of the year.

Check out our landing page or jump straight into the repository:
- Website: https://skilful-fox.com
- GitHub: https://github.com/Skillful-fox-Studio/grey-fox-community