✳ ONE SMALL SHOP · FOUR LINES · EVERY BUILD IN PUBLIC
LuckeySystems is a one-person engineering shop that builds software with AI agents and publishes the proof — prompts, source, and all. Four lines come out of it: Jarvis AI, a voice copilot you buy once and own; an arcade of 62 browser games, 30 of them written end-to-end by frontier models with the prompts on file; Athena, the free local-first toolkit that does the work; and Solutions, the same machinery pointed at small businesses.
The same machinery sells four ways. One product you buy, one arcade you play, one toolkit that's free, and one practice you hire — with a foundation being formed to steward the open parts when they outgrow me.
VOICE · COPILOT FOR CODING AGENTS PROPRIETARY
Speak your intent and Jarvis drives Claude Code, Codex, Aider, and OpenCode — dictating tasks, steering sessions by name, approving or interrupting runs hands-free, and reading results back to you. By default, raw microphone audio and webcam frames stay on your machine — only conversation text reaches the model provider you configured. Realtime speech-to-speech, such as Grok Voice, is an opt-in exception that sends audio to that provider under your own account.
THE FOUNDING 100 — EVERY STEP IS PERMANENT
The first 100 licenses are numbered and priced in ascending steps, starting at $49 for waitlist members. When a step sells out its price never comes back. From customer 101 the standard lifetime price is $999. Every tier is the same product and the same lifetime terms — only the number on your license changes. See the current step and what's left ↗
Every game here runs in a browser tab and costs nothing. 30 of them are the point: a frontier model was handed one prompt and shipped verbatim, no human edits and no second takes, with the prompt, model, and reasoning level published next to the game. It's a benchmark you can play — and the loudest evidence that the way this shop builds software actually works.
30 games. One standardized prompt, one model, shipped exactly as written — no human code, no retries. The prompt and model are published beside every entry.
32 games. Built in this shop with AI in the loop and a human steering. Bigger, more finished, and honestly labelled as such.
Opening later. A paid, reviewed tier for games you built and want hosted — sandboxed, credited, and held to the same provenance rules.
Most AI tools wake up with amnesia. Athena is the local-first toolkit that fixes that — memory, recall, and orchestration that stay in files on your disk. It's what the games and the client work get built with, and it will always be free to run. Athena Code and Athena Loops are MIT; the rest is source-available.
curl -fsSL https://raw.githubusercontent.com/luckeyfaraday/athena-code/main/scripts/install.sh | bash
A local command room for AI coding agents. One Electron app embeds native terminals for Codex, OpenCode, Claude Code, and Hermes, resumes sessions already on disk, generates bounded project handoffs, and keeps recall context ready for the next fresh agent. MCP bridge included, so your agent can drive the workspace itself.
A terminal AI coding agent with persistent local memory. It saves stable
facts and decisions, automatically recalls what's relevant each turn, and
searches across every past session — without shipping your context to anyone.
A memory-focused fork of OpenCode with native memory_write,
memory_read, and session_recall tools.
A deterministic harness for multi-agent work. Loops decomposes a goal into subtasks, runs worker agents in parallel, then gates the results through a reviewer that repeats until they pass. The control flow is real code, not a prompt — and the same loop drives whichever backend you bring: Claude, Codex, OpenCode, or aider. Ships as an MCP server and a CLI.
Talk to any text field on your desktop. A floating always-on-top widget that records, transcribes locally with faster-whisper, cleans up spoken punctuation, and types the result into terminals, coding agents, browsers — whatever has focus. The local-first answer to cloud dictation tools like Wispr Flow.
Athena Code freezes one deterministic memory snapshot per session, recalls related facts on every turn, and indexes prior conversations — its own and your local Claude Code, Codex, opencode, and Hermes histories — with SQLite FTS5.
Memory, recall, transcripts, and audio live in files on your disk. The hot paths need no backend service and no account. The cloud is an option, never a dependency.
Everyone says AI built it. Here the prompt, the model, and the reasoning level are published next to the thing itself — and where a human steered, it says so on the label.
Codex, Claude Code, OpenCode, Hermes — bring whichever agent you already pay for. Athena is the workshop around them, not another walled garden.
The tools are free and stay free. The one product that costs money is bought once and owned forever — no subscription, no seats, no upsell to the version that actually works.