A one-person AI tooling studio

Build it.
Understand it.
Share what works.

This is a workshop, not a company. I make things, figure out how they actually behave, and share the parts that turn out to be useful - mostly around terminals, AI systems, and how to bring them together without giving up control.

Meet ridinCLIgun Visit the Garden

security > privacy > cost > usability

⬡

Build and share

Working tools for a concrete problem. First up: ridinCLIgun, a terminal copilot that helps, warns, but never takes control.

◎

Understand and write it down

Notes on AI safety, usability, agent-assisted workflows, and what is actually useful.

◇

Safe and open

Security first. Privacy second. Cost third. Usability follows from those constraints, not the other way around.

first product

ridinCLIgun

>_ridinCLIgun

A terminal tool for people who want to keep their shell, but occasionally want a quick and reliable second opinion before they press Enter. Building with agent systems taught me something simple: let the agent explain, run it yourself - that's how it sticks. ridinCLIgun was born in exactly that moment.

v0.4.7 shipped human in the loop local-first variant-aware help room to grow

Product page and install

See ridinCLIgun Open the product page

Install on macOS via Homebrew brew tap inference-garden/ridincligun brew install ridincligun Then run ridincligun. Local warnings work out of the box. For AI support, insert your API key. Source install via GitHub stays available as fallback.

Useful today, still growing.

Current build: variant-aware offline help, typo detection, language-aware AI, and review history are already working.
Human in the loop: ridinCLIgun advises. It never presses Enter for you.
Room to grow: the next steps lean toward teaching, workflow support, and supervising agents - without taking control away from the operator.

zsh - ridinCLIgun v0.4.7

Inference-Garden/ ridinCLIgun on master MU is v0.4.7

> curl -fsSL https://docker.com | sh

DANGER

Downloading and executing code directly - you can't inspect it first.

Download the script first, review it, then execute:
curl -O script.sh URL && less script.sh && bash script.sh

curl

Transfers data from or to a server.

Example commands from the offline catalog:

curl https://example.com

curl -L --location -D- --dump-header - https://example.com

curl -O --remote-name https://example.com/file.zip

AI: on Secret: off Version: v0.4.7 Ctrl+G -> help zsh

/garden/ - what's growing here

What's growing here

A few subjects I keep coming back to. The English Garden now has two articles and a companion glossary, so there is finally a proper path into the rabbit hole.

Agentic Coding and Security

The first English Garden article is live: a four-layer risk model for building safer code with coding agents.

Read part 1

Glossary

A companion sheet for quick definitions around coding agents, trust boundaries, prompt injection, and security checks.

Open glossary

Lethal Trifecta and Prompt Injection

Prompt injection, untrusted input, the Rule of Two, and the design moves that keep an agent's working context clean.

Open article

/garden/ is not a blog. it's a garden. it grows.

how I think about personal AI

Good tools make the user stronger, not more dependent.

Good tools make the user stronger, not more dependent. They explain what they do, and let the person learn and decide. They favour security first, privacy second, cost third, usability fourth - not because usability does not matter, but because it's the result of the before and the order decides whether the tool serves the person or the other way around. They do not promise more than they deliver. And they are fun - because getting things to actually work is one of the best feelings there is.

There is something of a civic duty in this: those of us who can figure these systems out should help others run their own agent setups safely, sovereignly, and affordably - participate in the discourse - and help make that the norm, not the exception.

inference-garden