The vault is open.
The 19Keys community archive — take the code, build your own.
Every project below is yours to clone, copy, and rebuild. Nothing here carries a live secret — each repo is scanned before it's published.
New here? Start with the syllabus →
START HERE — the syllabus: 21 steps, 6 phases, zero to running your own agents. Read this first.
LLM prompt compressor with an honest eval harness — 75 tests, measured savings, make it your own.
Make your agent email you first: daily digests and alerts via trigger, agent compose, and a safe stdlib sender.
Easy setup for AI agents: Claude Code, Hermes Agent, and OpenClaw (Clawd) — three tiers, copy-paste commands.
Cameo-style 'Request a Key' booking page template — single file, black+gold, deploy anywhere static.
The build-your-own-x tutorial index wrapped in a session-restartable coding-agent harness.
Autonomous chief-of-staff agent: heartbeat scheduling, cost guards, revenue dashboard. Stdlib-only Python.
Step-by-step playbook to revive a slow MacBook: diagnose swap thrash, find space hogs, clean safely.
Playwright design forensics: screenshot any site + extract its ranked design tokens (fonts, colors, spacing).
The motion and design layer: GSAP, Lenis, Motion, thinking-orbs, transitions.dev, react-bits, vanta, taste-skill, ponytail — with two working demos you can run in two minutes.
Curated map of genuinely free AI: models, local tools, voice, images, learning — with honest fine print.
Four marketing agents with real verification gates: funnel psychology, copywriting, email, CRM automation.
The skill that makes coding agents reliable: harness your repos with state, verification gates, and session handoff. One-command install.
Join Higgsfield Academy + set up the Higgsfield MCP so your AI agent generates images, video and audio directly.
The HIGH LVL video render tools: luxury grades, quote cards, X-post cards, screenplay clips. ffmpeg + Python.
Local-first voice assistant loop: Chatterbox voice clone on Apple Silicon with elevenlabs/say fallback chain.
Run AI fully local and private: Ollama, model sizing by RAM, local voice, wiring into agent workflows.
Set up Obsidian as your second brain and wire AI into it: vault structure, sync, Claude Code integration.
The 3-tier orchestration pattern: expensive model orchestrates and verifies, cheap models execute.
The external open-source repos in the 19Keys stack and why each earned its place. Build, don't spend.
One shared notebook, one dictionary, one name list — typed claim store + measurement contracts so your AI stops guessing. Stdlib Python, zero installs.
What Claude Code skills are, how to install them, and how to write your own — with a working example.
Drop-in scaffold for a sovereign AI agent stack — clone and make it yours.
A scoreboard for every maybe — forecast ledger with named forecasters, strict grading, and computed calibration so track records are measured, not remembered.
Put all your machines on one private encrypted network: install, MagicDNS, Tailscale SSH, serve your dashboards to your phone.
BotFather to phone-buzz: create a Telegram bot, get the token and chat ID safely, and wire your agent to message you.
Absolute-beginner terminal guide, Mac + Windows: shell basics to installing an AI coding agent.
The real 19Keys stack, forensically inventoried: every CLI tool, app, MCP server, repo, and service — all clickable.
Build your own Voice Architecture Engine as a Claude skill: map your voice so AI writes like you, not like slop.
Embeddable black+gold ZIION readiness widget. Copy the snippet, drop it on any page.