sethserver.com Subscribe
A robot character with a banana-shaped yellow body and mechanical arms and legs stands center on a cyan background, surrounded by floating elements including TV monitors and speech bubbles displaying code snippets (</> symbols), along with geometric shapes like triangles and circles in black, white, and coral colors.

King Louie - My Cross Computer, Multi-LLM AI Assistant

By Seth Black • Updated: June 29, 2026

AI · 3 min read

Like this kind of writing? Get one email a week with notes on startups, AI, and the occasional strong opinion about Python: subscribe to the newsletter.

You can learn a lot about a tool by watching it fail hard enough to make you say, "What the crap!?"

That was me the first time I saw an LLM confidently make something up. This was the ChatGPT 3.5 era: tiny context windows, a few thousand tokens. Ask a reasonable question, get a completely fabricated answer delivered with total conviction.

My coping strategy has always been simple: RTFM and jump in. Manuals, docs, research papers, GitHub repos. I like digging. So when I started building King Louie, I built the thing I wanted sitting next to me while I dig.

King Louie is my cross-computer, multi-LLM assistant

King Louie is a desktop app where you bring your own API keys and talk to multiple model providers in one place. OpenAI. Anthropic. Gemini. Groq. Mistral. Ollama for local models. OpenRouter if you want a meta-router.

That part is straightforward. What makes it feel real is that it goes past chat, into the stuff you do right after chat.

  1. Ask a question
  2. Get an answer
  3. Copy it into a file
  4. Run a command
  5. Watch it break
  6. Paste the error back

King Louie shrinks that loop.

Smart routing: stop manually picking models

If you use LLMs a lot, you already have a mental map:

King Louie has rule-based routing. Keywords, regex, or slash commands. First match wins. If nothing matches, it falls back to your default.

/design  -> Claude Sonnet
/docs    -> GPT-4o-mini
/code    -> GPT-4o
/refactor|redesign -> Claude

I'm sick of "magic". I want fewer context switches.

Agents that can actually do the work

King Louie ships with agents and tools. The ones I reach for most (I built it, so I have opinions): bash, read, write, edit, grep, git, web search, web fetch, browser automation, cron scheduling.

This is where my "read the docs" habit turns into something useful. The model can read and write files with you, run commands, and show what happened.

I like having a code-explorer agent that crawls a repo and tells me what's going on without me opening 42 files. Then a code-writer agent can implement changes while staying inside the boundaries I set.

Tool approvals matter. I want the assistant to be helpful, not to run rm -rf because it misread a prompt. This is where building agents that fix their own bugs becomes practical: agents that capture terminal output, diagnose failures, and retry with corrections instead of just executing blindly.

Mesh networking: my laptop can boss my desktop around

This part feels futuristic, but it came from a practical problem.

King Louie instances can pair over a secure mesh network. Each instance has an identity (Ed25519 keys + TLS cert). You pair with a short code. After that, you can dispatch work to another machine.

If my laptop is lightweight and my desktop has the GPU (or just has the repo checked out), I can do:

Chat plus remote execution, without hopping between SSH terminals all day.

Channels: the assistant can live where your team lives

King Louie also lives in your team chat (Telegram, Discord, Slack), with the same routing and agents as everywhere else. You can gate it with mentions and allowlists so it doesn't reply to everything and get itself banned.

Why I built it

I have too many ideas. That's a logistical problem.

I keep thinking about smaller models on smaller hardware. AI on the device. Manufacturing automation with small robots and small chips. A Jetson strapped to something that moves and does work.

King Louie is the software side: an assistant that reads the manual with me, runs the commands, and hops across machines.

And if it makes something up? I'll catch it. Been doing that since 3.5.

Check it out at https://github.com/the-banana-tool/king-louie. Builds for Mac, Windows, Linux.

-Sethers

Share this post
Newsletter

One email, once a week.

Notes on databases, systems, and the occasional strong opinion about Python. No spam, unsubscribe anytime.

Seth Black
Written by

Seth Black

Engineer and founder based in Texas. Writes about databases, AI, and running things in production. Embeds in small teams as lead engineer.

More from AI

View all →
AI

How to be a Good Farrier: What the Car Teaches Us About AI and Your Job

Nov 10, 2026
AI

Principle of Least Privilege in the AI Age

Apr 14, 2026
AI

How to Actually Secure MCP (Model Context Protocol)

Apr 14, 2026