software lab

YNFRA

Software built for
a specific purpose.

Software built for specific problems. Prototyped fast, validated early, shipped when ready.

Active work Get in touch
$ ynfra run → stage: early → approach: build · validate · ship → mode: product lab ❯

Active Work

Real projects. No filler.

A small set of real software and infrastructure projects. Public items are shown when useful and ready; internal systems stay private.

live

ynfra.eu

Public website and deployment pipeline for YNFRA.

ynfra.eu
internal

YNFRA Infrastructure

Private self-hosted infrastructure for development, Git, deployment, routing, and local AI workflows.

private
internal

AI-assisted development setup

Local workflow using Codex, Claude Code, Obsidian, Paperclip, OpenJarvis/Ollama, and server-side OpenClaw.

private

Featured Tool

YNFRA Project Launcher

v0.5.10 — dogfood

Local-first CLI for opening, documenting, and managing real project work sessions.

The problem

Context switching between code, notes, terminals, browser tabs, and project files creates friction before real work starts. Every session begins with ten minutes of reassembling state from memory.

The solution

One CLI command opens the project environment, creates a session note, launches configured open actions, checks project health, and supports safe Codex workflows — all from a declarative JSON pack, no hidden automation.

Core principles

  • local-first — no cloud sync, no background services
  • Windows-first, Python stdlib only
  • private configs stay in projects.local/
  • open_actions are declarative, never arbitrary shell commands
  • explicit safety boundaries on every launch path
  • Obsidian integration via registered vault registry
  • Git/Gitea as source-of-truth workflow
ynfra — project-launcher 
██╗   ██╗███╗   ██╗███████╗██████╗  █████╗ 
╚██╗ ██╔╝████╗  ██║██╔════╝██╔══██╗██╔══██╗
 ╚████╔╝ ██╔██╗ ██║█████╗  ██████╔╝███████║
  ╚██╔╝  ██║╚██╗██║██╔══╝  ██╔══██╗██╔══██║
   ██║   ██║ ╚████║██║     ██║  ██║██║  ██║
   ╚═╝   ╚═╝  ╚═══╝╚═╝     ╚═╝  ╚═╝╚═╝  ╚═╝

project-launcher · v0.5.10 · windows-first · local-first

Key commands

ynfra open <project> open the full project environment
ynfra start <project> --task "..." create a session note and begin work
ynfra finish <project> record outcome, blockers, next steps
ynfra doctor check environment and project health
ynfra validate --all validate all project packs against schema
ynfra codex <project> --task "..." --save-prompt capture a Codex prompt locally
ynfra add-action <project> append a declarative open action
ynfra remove-action <project> remove an open action safely

Demo flow

PowerShell — ynfra demo
PS> ynfra doctor
PS> ynfra open --dry-run bell
PS> ynfra start bell --task "Improve research overlay"
PS> ynfra codex bell --task "Improve research overlay" --save-prompt
PS> ynfra finish bell --note "Implemented first fix" --next "Test in browser"

dogfood This tool is being used internally before public screenshots and polish are added. Presented here as a real working YNFRA portfolio project.

Approach

Research. Prototype.
Validate. Ship.

Consistent process for internal tools and experiments alike. AI compresses the early cycles — research, scaffolding, iteration — keeping focus on the actual problem.

Active development

Experiments are developed privately and published when they're usable.

01

Define the problem first.

Understand what is actually broken before writing code. AI-assisted research maps the use case, existing tools, and real friction points quickly.

02

Build the smallest useful version.

A working prototype in days. Enough to test the core assumption — not enough to waste time on the wrong thing.

03

Validate with real use.

Does it solve the problem? Is the friction acceptable? Answered with a working build, not a mockup or a survey.

04

Improve with intention.

Every feature added after validation has a clear reason. Complexity is earned, not assumed.

05

Automate and deploy.

Repetitive logic, deployment, and data processing are automated early. What ships should work without constant attention.

Principles

How, not just what.

These apply to every project — from a two-day prototype to a long-running system.

Useful before impressive.

A tool that solves a real problem is worth more than one that looks like it might. Optimizing for utility, not presentation.

Validate before scaling.

Build the minimum that proves the idea works. Expansion comes after the core is proven — not before.

Automate what repeats.

If a task is manual and repetitive, it is a target. Automation is a standard part of how things get built, not an optional feature.

AI is leverage, not the product.

AI accelerates development, research, and iteration. Infrastructure for moving faster — not something to sell.