If you’ve been using AI agents like Claude or GPT to build software, you’ve likely felt the “blind spot.” You give a prompt, the AI writes code, and it looks okay, but is it actually good? This is where Sentrux steps in.
What is a Feedback Loop?
In software development, a feedback loop is the process of writing code, measuring its quality, and using those results to make improvements. For human developers, this usually involves code reviews or running tests. For AI agents, it’s about “recursive self-improvement.”
The Danger of the “Broken” Loop
Without a tool like Sentrux, AI agents operate in a vacuum. They can generate thousands of lines of code, but they lack a sensor to tell them if that code is becoming a “spaghetti” mess.
This is dangerous for developers for three reasons:
- Technical Debt Explosion: AI can create complex depth and redundancy faster than any human, leading to unmaintainable projects.
- False Confidence: An agent might report a successful build while the underlying modularity and acyclicity scores are plummeting.
- The Token Trap: When an AI doesn’t know exactly what is wrong, it guesses. It tries multiple broad fixes, re-reads the entire codebase, and outputs massive blocks of redundant code. This consumes a massive amount of tokens, spiking your API costs without actually improving the code quality.
How Sentrux Solves It
Sentrux acts as the “eyes” for your AI agent. It scans the code and provides a concrete Quality Score based on metrics like modularity, redundancy, and depth.
Instead of the AI guessing why the code feels “off,” Sentrux gives it a direct target: “Your redundancy score is 2016; get it under 1000.” This directness makes the agent more efficient, saves a fortune in token usage, and ensures that the recursive self-improvement is actually moving toward a better product, not just a larger one.
Quick Start
Sentrux is built in pure Rust and works across macOS, Linux, and Windows. You can get started with a simple brew or curl command:
- macOS:
brew install sentrux/tap/sentrux - Linux/Windows: Use the curl installers available on GitHub.
Stop letting your AI agents code in the dark. Give them the sensor they need to build better, leaner, and cheaper software.
License
The app is a free and open-source that is released under the MIT License.
