My technical write-ups. Easier on the eyes than a GitHub README.
I wired two local AI agent harnesses to a real router with one write tool. Here is what decides whether that write runs, and the three guardrails that held.
Run a free AI model on your own hardware, feed it real configs with zero redaction, and dodge the 4 silent failures that make local AI look broken.
My AI agent audited three live Cisco devices, found a real CVE, and wrote the fix. My server-side boundary would not let it apply the change without me.
I gave Claude read-only SSH access to my home lab with a custom FastMCP server. It diagnosed an OSPF problem from live device output in about 30 seconds.
How 15 AI prompts for config generation, troubleshooting, ACL audit, and documentation came together. The 4-piece structure, the testing, the limits.
One Python script takes you from SSH to a finished, AI-generated network runbook. No copy-paste, no browser chat, no manual steps.
Use Netmiko and AI to turn raw device output into structured, readable network documentation automatically.
A rapid triage method using tools you already have to prove your network is innocent in 60 seconds.
