This is a submission for the OpenClaw Challenge.

ClawNet — AI Security Terminal Powered by OpenClaw

What I Built

We all have a habit of downloading open-source GitHub repos, random tools, scripts, cracked installers, and trying new software without really knowing what they’re doing in the background.

The problem is simple: unless you already have strong endpoint security running, you usually have zero visibility into what these processes are doing on your machine.

That’s where ClawNet comes in.

ClawNet is a Windows-native AI-powered interactive security terminal that continuously monitors live network connections, maps them to running processes, detects suspicious behavior, and uses OpenClaw to explain whether something is safe, suspicious, or critical.

Instead of just showing “unknown.exe connected to 185.xxx.xxx.xxx,” it tells you:

• why it’s dangerous
• whether it looks like malware or C2 beaconing
• whether the binary is suspicious
• whether you should kill the process or block the IP

It acts like an intelligent network watchdog between the user and hidden threats.

Core value:

Detection → Analysis → Explanation → Response

Not just alerts. Actual decisions.

How I Used OpenClaw

OpenClaw is the brain of ClawNet.

The system first monitors active TCP/UDP connections using psutil, maps each connection to the exact process running on Windows, checks file paths, suspicious ports, VPN state, GeoIP location, and process legitimacy.

Then OpenClaw takes over.

OpenClaw powers:

1. AI Threat Classification

Every suspicious connection is sent to OpenClaw (GPT-4o-mini engine) where it classifies it into:

• SAFE
• SUSPICIOUS
• CRITICAL

It returns structured JSON with:

• threat level
• reasoning
• action recommendation

Example:

{
  "level": "CRITICAL",
  "reason": "Unsigned binary connecting to high-risk foreign ASN",
  "action": "kill_and_block"
}

This removes false positives and turns noisy logs into useful decisions.

2. Persistent Threat Memory

I added memory using Supermemory SDK with local JSON fallback.

This means ClawNet remembers:

• suspicious IPs
• repeated malicious processes
• previous kill/block decisions
• flagged binaries
• user approvals/rejections

Without memory:

“node.exe looks suspicious”

With memory:

“node.exe connected to this IP 3 times this week and was previously marked suspicious”

This dramatically reduces hallucinations and improves context-aware threat analysis.

3. Telegram Remote Alerts

Critical threats trigger Telegram alerts instantly.

So even if I’m away from my machine, I can get:

• threat alerts
• severity score
• reason for detection
• recommended action

This makes it useful beyond a local terminal and closer to a lightweight personal SOC system.

4. Copilot Security Mode

There’s also an interactive “Copilot Mode” where I can ask:

Why is node.exe connecting to this foreign IP?

and OpenClaw explains it in plain English like a security analyst would.

This makes it useful for both technical and non-technical users.

Demo

GitHub Repo

Project Repository:
https://github.com/rajarshidattapy/clawnet

Example Workflow

unknown.exe → foreign IP → unsigned binary

↓

OpenClaw analyzes behavior

↓

CRITICAL threat detected

↓

Telegram alert sent

↓

Suggested action:
kill process + block IP

Live Terminal Experience

ClawNet runs as an interactive terminal showing:

• active network connections
• process names
• remote IPs
• risk scoring
• AI-generated verdicts

with autonomous response recommendations.

What I Learned

The biggest thing I learned:

Security tools fail when they only show data.

Users don’t need more logs.
They need decisions.

Raw monitoring is easy.
Context-aware threat reasoning is hard.

That’s where OpenClaw became genuinely powerful.

Another major realization was how important memory is.

Without persistent memory, AI keeps re-analyzing the same threats like they’re new. With memory, it starts behaving like an actual security system instead of a stateless chatbot.

I also learned that building autonomous recommendations is far more useful than simple detection.

People don’t want:

“This looks suspicious”

They want:

“Kill this right now.”

That shift changes everything.

ClawCon Michigan

I did not attend ClawCon Michigan, but building ClawNet made me realize how powerful personal AI infrastructure can become when it moves from “assistant” to “operator.”

OpenClaw feels less like prompting an LLM and more like building an actual intelligent system.

That’s the future I’m interested in building.

ClawNet is not just another monitoring tool.

It’s an AI security layer for the open-source era—where we install first and regret later.