muthuishere

Most local LLM setups are not portable. This bundles model, embeddings, chunks and metadata into a single export. The exported package runs fully in the browser without internet or install. Trying to solve reproducibility and deployment in restricted environments. https://github.com/muthuishere/offline-llm-knowledge-system �

I like using Excalidraw for diagrams, especially the hand-drawn style. But I don’t like starting from a blank canvas. I usually spend 10–15 minutes placing boxes and arrows before I even get to the actual thinking. I tried some existing Excalidraw MCP/agent skills, but they didn’t work well for me. No image export, no animation, and the style didn’t match what I wanted. So I built a small agent skill for my workflow. It generates a hand-drawn diagram, an animated version, and opens it in a hosted Excalidraw (GitHub Pages) where I can edit it further. I still tweak things manually, but this gives me a starting point instantly. Repo: https://github.com/muthuishere/hand-drawn-diagrams

https://deemwar.com/. Helping startups to grow

I run a very small VPS to host demos for my open source work. Traffic is minimal (maybe 10–20 users), so I assumed no one really cared.

After checking the logs, I realized that assumption was wrong.

Even with almost no real users, the server was constantly scanned: SSH brute-force attempts, HTTP probing for .env, AWS credential paths, and random endpoints.

Nothing broke, but it was clear I wasn’t really watching.

I explored a few options and ended up using CrowdSec. At first it felt heavy and not very friendly for a Docker + Kamal setup, but after some trial and error I got it working and automated.

I wrote up what I learned:

what SSH and HTTP logs actually look like on a “no-man’s” VPS

why repeated 404/403 probing matters

why temporary bans are safer than permanent ones

how I automated the setup so it’s repeatable

Article: https://muthuishere.medium.com/securing-a-production-vps-in-...

Video walkthrough: https://youtu.be/hSiMfbJ4c0Q

Automation / source code: https://github.com/muthuishere/automated-crowdsec-kamal

Sharing this in case it helps someone else running a small public server who assumes it’s too boring to be attacked. Happy to answer questions or hear how others handle this.

Docker Desktop on macOS often sits in a VM that constantly sucks CPU. For years I tolerated it—until today, when running just two Node.js apps + one Python app pegged my system, froze the UI.

I switched to Colima and pointed docker context use colima. Then added Portainer for UI (docker run -d -p 9000:9000 ...).

With the same compose setup, Colima usage dropped to ~0.2%

Portainer provides containers/images/console UI

All while keeping the Docker CLI workflow

Full write-up (with screenshots) here: https://muthuishere.medium.com/docker-desktop-colima-portainer-my-mac-finally-breathes-b70d82d6cf0f

You build tools. Shell scripts. Docker commands. Things you want to run from anywhere.

But every OS has different rules:

Mac/Linux: edit .bashrc, .zshrc, mess with PATH, source it Windows: create .bat, set up %PATH%, hope PowerShell respects it

You waste time managing commands instead of building things.

LNB solves that in one line:

npm install -g lnb

Then:

lnb alias build "docker run -v $(pwd):/workspace node:18 npm run build" build

Or:

lnb ./my-tool my-tool --help

Use cases:

Make any binary globally accessible

Wrap long dev commands into short aliases

Share consistent workflows across your team

Example:

lnb alias deploy "docker-compose -f prod.yml up -d" lnb alias logs "kubectl logs -f deployment/app" lnb alias test "npm test -- --coverage"

Cross-platform:

On macOS/Linux → symlinks or shell scripts in /usr/local/bin

On Windows → .cmd wrappers in %USERPROFILE%\bin

Works in Bash, Zsh, CMD, PowerShell, Windows Start menu search

Built for:

Developers who build CLIs or run local scripts

Teams who want consistent dev commands across machines

Anyone tired of tweaking PATH manually

Links: GitHub: https://github.com/muthuishere/lnb

NPM: https://www.npmjs.com/package/lnb

Encrypt in Java, decrypt in the browser (or seed JS-encrypted payloads into Java). Sounds straightforward. It’s not.

You hit landmines like:

Mismatched IV/tag defaults (Java Crypto API vs WebCrypto)

Conflicting key formats (X.509 vs JWK vs raw)

Varied Base64 implementations and encoding subtleties

I spent hours chasing these. So I made Dual-Crypt, a fully working example:

AES-256-GCM + RSA-2048, wired correctly across Java ↔ JS

Uses spki for public key, pkcs8 for private key, 12-byte IV, 128-bit tag

Copy-paste ready code and a live demo

Read how I hacked it into existence, and try it yourself: https://muthuishere.medium.com/dual-crypt-cross-platform-encryption-spring-boot-react-that-actually-works-c713ecdbb89c Live demo here: https://dualcrypt.muthuishere.site