An OS-level security layer and telemetry pipeline designed to safely sandbox and monitor any autonomous AI agent, built for HackMerced.

We didn't just fix or fork an existing AI codebase. Because our firewall wraps Operating System system calls via LD_PRELOAD, our software can protect against destructive actions from any autonomous AI agent on the market—OpenClaw, AutoGPT, Devin, etc.

C/C++ Core Systems (Yash): The LD_PRELOAD library hooks into unlink, rm, and open to instantly teleport files rather than delete them, operating in $O(1)$ time.

Telemetry Pipeline (Souradeep): A Python wrapper streams all AI logs (stdout) over Wi-Fi directly to a Splunk Cloud index via HTTP Event Collectors.

MLFQ Rules Engine (Souradeep): Splunk SPL identifies malicious intent and bounces a Webhook back to our Edge Hardware.

Kill Switch: A local Flask server receives the webhook and triggers our C++ Dispatcher to execute a fast kill -9 on the rogue AI.

Real-time Dashboard (Harshith): A React UI hooked into Firebase displays live intercept numbers and neutralized threats.

Clone this repository (which contains NO third-party AI code, making our codebase 100% pure).

Build the Edge Docker container (which will dynamically download OpenClaw to serve as the target test subject):

Watch the terminal as our C++ Hook intercepts AI anomalies!

We require vite for this next step.

Launch the dashboard: