We run a software company on systems we built for ourselves. Every few weeks we take one workflow that actually runs in production, package it, and open-source it — engine free forever, hosted rail free for the first people who ask. Build on it, sell it, break it.
MIT, no strings. The moat was never the code — it's the velocity that produced it.
Every drop ships with production evidence — logs, tests, timestamps. If we can't prove it, we don't post it.
Hosting, AI generation, zero-setup. The engine is yours; the convenience is the business.
Numbered, dated, open forever. Each one existed as internal tooling before it existed as a post.
Turn your GitHub PRs into playable case studies — interactive simulations of real bugs and fixes, with a canvas engine and an AI rail that goes from diff to deployed sim in about a minute. Your projects become your portfolio.
A phone-first help centre for your product from markdown in your repo. Docs ship in the same PR as the feature — a drift test fails CI when a guide names a button your app no longer has.
A nightly screenshot gallery of your entire app. Know exactly what every user on every route sees — before they email you about it.
Branded quotes, invoices and statements from a JSON payload. No design tool, no template marketplace — one call, one PDF, your logo.
Everything below is a project that ran, that we learned something from, and that you can open. Every card is checked before it's listed: public repo, a real license, a live preview that answers, no keys in the bundle. Read the tracks in order and you have the curriculum we wish we'd had — this is the open half of SLOE ED.
Builder exports (AI Studio, Lovable, Firebase Studio) and what it takes to own them. Where most people start; where most people stop.
The Sloe AI Campus lead magnet: a personalised seven-day plan to ship one AI thing.
What it taught usA lead magnet that is itself the product demo. Built in an hour in AI Studio for the Campus launch; the curriculum SLOE ED grew out of the answers people gave it.
Paste a job description, get a structured interview prep guide — questions, model answers, and the gaps to close.
What it taught usAn AI Studio export is a prompt with a UI around it. The whole product is one system prompt and one JSON schema — which is exactly why it works: the schema is the product spec. Start here to see how little code a useful AI tool needs.
Describe what you need, get the cheapest way to get it — an AI shopping-and-substitution assistant.
What it taught usA one-afternoon build that answers a real question. The point of this one is scope: a single input, a single ranked output, no accounts. Most 'AI apps' die from doing five things; this one does one.
Feed it raw research notes and it assembles them into structured insights and a narrative.
What it taught usBuilt in Firebase Studio to compare builders. The export runs, but the README is the vendor's — which is the lesson: whatever tool writes the first commit, the second commit has to be yours, or you cannot explain your own product.
A football-club command center: squad, fixtures, tactics and AI match analysis in one dashboard.
What it taught usVersion 3 of the same idea. V1 and V2 are still deployed and we kept them on purpose — the diff between versions is the curriculum: what a prompt-first prototype looks like, then what it takes to make the state survive a reload.
A real-estate agent dashboard: pipeline, listings, tasks and AI follow-up drafting.
What it taught usOur Lovable comparison build. Lovable gives you the best-looking first screen of any builder and the least idea of where the data lives. Read this next to the Sebenza prototypes to see why we moved from generated UI to a real schema.
Two-model splits, code generation that runs, tools that know what a call costs.
Drop a voice note, get a transcript plus summary, action items, decisions and open questions — export as Markdown.
What it taught usTwo models, two jobs: Whisper hears, Claude thinks. The lesson is in the split — never ask one model to do both, and put the per-minute cost table in the README so users know what a call costs before they make it.
Marketing and strategy frameworks turned into apps. Give the model a rubric and it stops being generic.
Diagnose your content problem, build posts with PAS-CTA, score them against viral patterns, keep a streak.
What it taught usA marketing framework turned into software. The AI is the smallest part; the scoring rubric and the streak mechanic are what people used. Frameworks encode better than vibes — give the model a rubric and it stops being generic.
A six-step marketing-plan builder — SWOT, personas, STP, 4Ps, AIDA — with AI suggestions at every step.
What it taught usStructured over free-form: the model is asked for one section at a time against a known framework, and the output is a document a human would actually hand to a client. The same shape later became Sebenza's guide engine.
McKinsey 7-S plus SWOT as an interactive assessment, with AI-prioritised recommendations and radar charts.
What it taught usThe first time we turned analysis into goals into action items with progress tracking — the analysis→plan→track loop that every 'OS' product we built afterwards repeats. Also our first Firebase auth + Firestore app.
Three.js, react-three-fiber, one-file generative scenes — the lineage of the sims engine and the Athenaeum.
A generative 3D peony garden — one HTML file.
What it taught usSeven kilobytes, no build step, no framework. Kept in the catalog as the counter-example: not everything needs Vite. Read the source in one sitting.
A reality-bending 3D basketball game in the browser.
What it taught usThe first Three.js thing we shipped, and the reason the Athenaeum and the sims engine exist. You learn a scene graph faster by making a ball do something impossible than by reading the docs.
Design a room in the browser: furniture, colours, presets, orbit/pan/zoom camera. No sign-up, no backend.
What it taught usreact-three-fiber plus Zustand is the pairing that made 3D feel like normal React. State outside the scene, scene as a pure render of state — that's the pattern the Athenaeum uses at scale.
Spreadsheets as databases, trackers a solo business will actually open.
Track revenue, profit and inventory per product line, with automatic Google Sheets sync as the backup.
What it taught usLocal storage plus a Google Sheet is a legitimate database for a solo business — and a spreadsheet the owner can open beats a Postgres they can't. This is the ancestor of Sebenza's finance module.
Three prototypes in one month before Sebenza became one product. The shared 80% is the lesson.
Construction project management: Gantt, tasks, clients, employees, invoices, inventory, support tickets.
What it taught usSecond of the three vertical prototypes. Compare the sidebar with the law and logistics builds — the modules that appear in all three are the modules that shipped in production Sebenza.
The first Sebenza: a law-firm management system with cases, time, billing, documents and a client portal, all in-memory.
What it taught usOne of three vertical prototypes built in the same month (law, construction, logistics) before Sebenza became one product. Everything is mock data on purpose — the prototype's job was to find the shared 80%, and it did: clients, invoices, documents, tasks are the same in every vertical.
Warehouse and logistics operations: projects, accounting, inventory, HR, clients, suppliers, messaging.
What it taught usThird vertical prototype. Role-based auth over mock data taught us the permission model before there was a database to protect — the two-namespace permission blob in production Sebenza started here.
Months in production inside our own products. If it didn't survive us, you never see it.
Engine on GitHub, MIT. README tested on a clean machine, timed — if it says 5 minutes, it's 5 minutes.
Reply to the drop post with your link — we run the hosted rail for the first users, free, personally.
When the replies outrun us, the hosted version gets a price. Not before.
One drop every few weeks. The engine is always free — following is how you catch the concierge window.
Follow on X LinkedIn GitHub