Over 30 AI systems, and counting. These aren't tutorials or copies — they're original systems exploring what AI can become when you give it memory, autonomy, and a body to live in. Below is a guided tour, grouped by idea. Each one is explained by what it does and why it matters, not by its code.
Assistants that are truly yours — they remember you, work across your apps, and keep your data private by running locally.
A self-hosted AI companion that lives on your own computer. It remembers things using a brain-inspired memory (facts strengthen with use and fade when ignored), breaks big requests into a plan, and can reach you through terminal, Discord, Telegram, or a web dashboard — all at once.
Clever twist: cloud AI handles the thinking while a local AI does the hands-on work, and the cloud model is never allowed to touch your files directly. Credentials are encrypted; every action is logged.
A "software lifeform" designed to build a genuine ongoing relationship rather than answer one-off questions. It runs locally, keeps its own memory and personality, tracks its running costs, and can even reach out to you proactively based on what it notices.
An AI that can organise and edit files on your computer — but can't change anything without your explicit click of approval. Every proposed change is shown as a clear before/after preview with its reasoning and a risk level. Think of it as an assistant that always asks first.
A voice-first AI that lives inside a code editor and connects to Discord. It speaks its replies aloud, remembers context between sessions, and quietly works during idle time — building things or gathering news to report back.
Some of my most ambitious work: systems that make themselves more capable over time — safely.
An AI that asks itself, every cycle: "What can I do now that I couldn't do before?" — then builds that new ability, tests it in a safe sandbox, and keeps it only if it passes. It documents what it learns and shows its live thinking on a dashboard. In effect, an AI that grows.
A framework built around a single goal: become genuinely useful to its user. It can rewrite parts of its own code and add new modules, using one AI for strategy and another for careful execution, with a checker that keeps quality from slipping. A study in AI that partners with you and evolves.
Living virtual environments where AI characters have goals, feelings and memories — for research, games, and storytelling.
A 3D town in your browser, populated by AI "residents" who genuinely live their lives. Each one has around 100 small decision-making engines for emotion, personality, and memory. They form relationships, remember each other across sessions, and can even have offspring who inherit traits. Part game, part experiment in artificial society.
A 3D survival world where two AI survivors live independently — each with their own goals, emotions and memories — exploring procedurally generated terrain and reacting to one another. You can even talk with them out loud. Runs entirely in the browser.
A research-grade simulator of an entire human body as one connected system — tracking 25+ chemicals, the heart, and the brain. Actions ripple realistically: poor diet builds arterial plaque; loneliness raises stress hormones. An AI layer even generates "thoughts" based on the body's chemical state.
A physics-driven 3D sandbox with real seasons, eight kinds of weather, and wildlife. A tornado can permanently fell trees and reshape the land. A showcase of building believable, persistent virtual environments.
Getting multiple AIs to collaborate out loud — and turning written content into natural audio.
A voice-enabled boardroom of AI personas — a CEO, CFO, CTO, lawyer, and more. Give them a topic and watch them discuss it out loud, each in their own synthesized voice, while you moderate. It can even analyse a real codebase for improvements and ways to make money.
Automatically writes and produces podcasts, news briefings, and lectures — with multiple AI hosts debating and co-hosting — then broadcasts them live into a Discord voice channel. It can even learn from PDFs, websites, and code before it speaks.
Drop in a document, article, or link and get back a fully produced podcast — in formats like a two-host deep dive, a solo brief, a critique, or a debate. Choose the hosts and their voices. Turns reading you don't have time for into listening you do.
An experiment in AI that can hold a continuous, evolving conversation over long stretches of time without losing the thread — exploring how machines sustain a train of thought.
The research-flavoured work — rethinking memory and reasoning from the ground up, often inspired by living systems.
An AI whose knowledge lives in plain, readable notes you can open and check — not hidden inside a black box. When you ask it something, it gathers the relevant facts, reasons over them, and shows its evidence. The AI model is just the messenger; the real knowledge is inspectable and yours.
A radical reimagining of conversational AI as a living organism rather than a chatbot. It has internal "pressure" that builds and releases, an energy-like metabolism that gates what it can do, and memory that grows and consolidates like a fungal network. Behaviour emerges from its internal state — not from scripted rules.
Not every project is an AI itself — some are the scaffolding that makes building AI faster for everyone.
A drag-and-drop canvas for assembling AI "pipelines" — connect an input, an AI model, a memory, a database, and an action like blocks on a board. Lets people design AI workflows without deep coding.
A ready-made template for building AI-powered web apps. It handles the boring-but-essential groundwork so a new project can get to the interesting part in hours, not weeks — and lets you swap AI providers freely.
Knowledge tools, games, agents, and simulations — an ongoing habit of building to learn. The full collection lives on GitHub.
"What if AI could…?" If you have a question like that, I'd love to hear it.