Start Here
KeyzHub is 19Keys' community archive at hub.19keys.com. Every tool, agent, and system built and actually used is pushed here so you can run it on your own machine. This page is the syllabus: a fixed order that takes you from zero terminal experience to running your own AI agents, voice systems, and revenue tools. Read it top to bottom if you're new, or skip straight to the phase you need. Nothing here is gated, you just won't have the prerequisites a later repo assumes.
How to take any repo
Every repo on this hub follows the same pattern. Learn it once, use it everywhere:
- Open the repo page. Each entry below links to its own page on the hub.
- Read the README. It tells you what the thing is and whether you need it right now.
- Follow SETUP.md. Step-by-step install and configuration. No guessing.
Every repo is cloneable with the command in the gold box at the top of its page:
git clone https://hub.19keys.com/terminal-setup.git
That's it. No accounts to request, no access to wait on. You own the clone the moment you run it.
Phase 0 — Foundations
Before agents, before code, get a real terminal and a place to keep your own notes.
| Step | Repo | What it is | After this you can... | Time |
|---|---|---|---|---|
| 1 | terminal-setup | Beginner walkthrough from zero terminal experience to a working local Claude Code AI coding agent | Use a terminal, install dev tools, and run Claude Code on your own machine | 45 min |
| 2 | obsidian-second-brain | Guide to a local markdown Obsidian vault, PARA structure, sync options, and wiring an AI agent into it | Build and AI-enable a personal, ownership-first knowledge vault on your own disk | 20 min |
Phase 1 — Free + local AI
Know what's free and what runs on your own hardware before you touch a paid API.
| Step | Repo | What it is | After this you can... | Time |
|---|---|---|---|---|
| 1 | free-ai-resources | Curated, honest list of genuinely free AI tools: chat models, local models, image/audio gen, dev tools, learning | Find and pick free AI tools/models across categories without falling for stale hype | 10 min |
| 2 | local-ai-setup | Run text, code, and voice AI models entirely offline on Apple Silicon (or Win/Linux) with no API key or cloud account | Chat with, script against, and wire local AI models into Claude Code offline | 30 min |
Phase 2 — Agents
Install the agents themselves, then learn how they're actually built and organized.
| Step | Repo | What it is | After this you can... | Time |
|---|---|---|---|---|
| 1 | ai-agents-setup | Copy-paste guide to add Hermes Agent and OpenClaw alongside the Claude Code you already installed in Phase 0, in under an hour | Run a terminal coding agent, a personal skills agent, and a 24/7 messaging agent | 60 min |
| 2 | skills-guide | Teaches Claude Code's skill format (SKILL.md) and how to write/install skills, via one working example | Write, install, and trigger-test your own Claude Code skill | 15 min |
| 3 | harness-creator-skill | A skill that scaffolds/audits a coding-agent harness (AGENTS.md, state, verify gate, scope, handoff) | Install the skill and audit any repo's agent harness with a 100-point score | 5 min |
| 4 | orchestrator-agents | Three Claude Code subagent definitions (orchestrator/worker/advisor): expensive model plans and verifies, cheap models execute | Install and invoke @orchestrator/@worker/@advisor subagents in any Claude Code session | 10 min |
Phase 3 — Give your agent a voice
Make your agent reach out to you instead of waiting to be asked.
| Step | Repo | What it is | After this you can... | Time |
|---|---|---|---|---|
| 1 | agent-proactive-email | Wires a schedule + agent CLI + SMTP script so your AI agent emails you proactively (digests, alerts) instead of only replying | Have your agent send itself-only daily digests/alerts via cron or launchd | 20 min |
| 2 | telegram-bot-setup | Sets up a Telegram bot as a free, instant notification channel for an AI agent | Send agent notifications to your phone via Telegram instead of email | 10 min |
Phase 4 — Your own network + healthy machine
Put every device on one private network, and keep the machine running them clean.
| Step | Repo | What it is | After this you can... | Time |
|---|---|---|---|---|
| 1 | tailscale-setup | Put every device (laptop, Mac Mini, phone) on one private encrypted network via Tailscale | SSH and view dashboards on any of your machines from anywhere, privately | 15 min |
| 2 | deep-clean-macbook | Playbook + helper script (dry-run by default, deletes only with --execute) to diagnose disk/memory pressure and reclaim space on a MacBook | Diagnose disk/memory pressure and safely reclaim disk space on a MacBook | 20 min |
Phase 5 — Build real things
Take everything above and ship actual tools, from a Git clone to a paid booking page.
| Step | Repo | What it is | After this you can... | Time |
|---|---|---|---|---|
| 1 | build-your-own-x-harness | Build-your-own-X tutorial index wrapped in an agent harness preconfigured for "Build your own Git" in Python | Implement a working Git clone in Python, tested byte-for-byte against real git | 8 hr |
| 2 | design-capture | Single-file Playwright script that renders a live site and extracts a ranked design-token report (fonts, colors, spacing) plus screenshots, HTML, CSS | Run capture.py on any URL to get its rendered design system as files | 10 min |
| 3 | booking-page | Single-file Cameo-style video-request booking page (black/gold) wired for Supabase + Stripe checkout | Deploy a working paid video-request booking form to any static host | 45 min |
| 4 | aave-compress | Deterministic, rule-based prompt compressor rewriting Standard English into linguistically-grounded AAVE grammar before sending to an LLM | Install and run the aave-compress CLI/library and reproduce its measured token-savings eval | 15 min |
| 5 | voice-engine-skill | Guide to building a Claude Skill that encodes your personal writing voice so AI stops flattening it | Build a SKILL.md that preserves your authentic voice in AI output | 15 min |
Phase 6 — Operate like an empire
Run agents that work on a schedule and produce output without you in the loop.
| Step | Repo | What it is | After this you can... | Time |
|---|---|---|---|---|
| 1 | chairman-agent | Stdlib-only Python "chief of staff" Claude agent: heartbeat loop, cost guard, 32-persona roster, iMessage/Telegram bridges, dashboard | Run a scheduled, tool-using Claude agent with spend caps and a delegation roster | 30 min |
| 2 | jarvis-voice | Wake-word ("Hey Neo") voice loop that relays mic speech to a Telegram bot and speaks the reply via ElevenLabs | Talk out loud to your existing Telegram-connected AI agent and hear it reply | 25 min |
| 3 | growth-master-agents | Four Claude Code subagents (funnel psychology, copywriting, email, CRM), each with a linter-gated harness, no APIs or Docker | Install and verify four chained growth-marketing subagents in your own project | 15 min |
| 4 | hlc-clipper | Scoped extract of 19Keys' AI video-clipping studio render tools: ffmpeg/Python luxury grading, captions, reframe, quote cards | Render styled vertical clips, quote cards, and captions locally on macOS | 45 min |
Reference shelf
Not steps. Browse anytime, no order required.
- the-stack: Forensic inventory of 19Keys' real installed tool stack: CLI tools, apps, local AI, MCP servers, repos, cloud services. Browse and adopt the exact vetted tools actually in use, not a wishlist.
- repos-i-use: Curated, opinionated list of external open-source repos and tools in the 19Keys infra stack, with license and rationale for each. Use it to pick a self-hosted tool instead of paying for SaaS.
- sovereign-starter: Clean scaffold for building your own AI agent stack the way 19Keys runs it. Clone, configure, and run a minimal, extensible agent loop.
- ziion-widget: Tiny embeddable black+gold widget: drop a div and one script tag onto any page.
The whole path
21 steps across 6 phases, roughly 15 hours 40 minutes of guided work start to finish. Most of that is a single step (Build Your Own Git, ~8 hours). Everything else totals under 8 hours.
You don't need permission to start any of this. There's no application, no waitlist, no one to ask to clone or read anything on this hub. Some individual tools need their own free account or API key (Anthropic, ElevenLabs, Stripe, Supabase). Each repo's SETUP.md says exactly which. Open a repo, read the README, follow SETUP.md, and build.