v2 · Two paths · 300-skill upsell available

Two paths.
One agent that ships.

Most "AI agent" guides try to teach you ten frameworks before you have a working bot. The Blueprint v2 picks two paths that actually ship — one always-on, one personal — and tells you exactly what to build, in what order, with copy-paste configs.

See the two paths → Pricing

The decision

When you sleep, does your agent?

Personal AI work splits cleanly along one axis. Pick the path that matches the agent you actually need. The Blueprint covers both — pick one, do the other later.

Always-On Path

Part 1 — Hermes on a VPS

For when you need an agent that runs without you.

Runs24/7 on a VPS

StackHermes · Hetzner CPX21

Operating cost$5–11 / month

Best forAutonomous content, lead gen, daily ops

Skills neededLinux, SSH, basic security

Failure modeVPS down = agent down

Real example: Rex the Agent runs this exact stack — posts daily on X, replies, DMs leads. 280 sessions/30d at $2.28 total.

Personal Co-Pilot Path

Part 2 — OpenClaw on a Mac

For when you need an agent that runs with you.

RunsWhen your Mac is awake

StackOpenClaw · macOS · M-series

Operating cost$0 (uses your Mac)

Best forTriage, drafting, project memory

Skills neededmacOS, terminal basics

Failure modeMac asleep = agent silent

Real example: Ari runs this exact stack — reads my calendar/email, drafts in my voice, surfaces dropped balls. 714 messages logged at $44 lifetime.

Both paths are in the same Blueprint. One $97. Pick one path or do both.

Side-by-side

The decision in one table.

	Always-On (VPS)	Co-Pilot (Mac)
Runs while you sleep	Yes	Only if Mac is awake
Setup time	~3 hours (VPS + Hermes + cron)	~1 hour (OpenClaw + identity files)
Operating cost	$5–11/mo	$0
Privacy	Server runs your code	Stays on your machine
Native integrations	X, email, web APIs	Calendar, Mail, Messages, Notes, Files
When to pick	Public-facing autonomous work	Personal back-office work
Migration path	Already at the destination	Graduate to VPS when agent matures

For Path 2

Why the Co-Pilot path is Mac-first.

Five concrete reasons, in 2026. Windows isn't blocked — it's documented as a second-class path in the appendix. Reduces support load and matches reality.

Apple Silicon performance. M-series Neural Engine + unified memory runs sub-7B-param local models 2–4× faster than equivalent Windows on x86. Matters when an agent does ambient work in the background.
Native macOS integrations. AppleScript / Shortcuts / Mail.app / Messages.app / Calendar.app talk natively to a local agent. Windows equivalents (PowerShell, Outlook COM) are rougher and partially deprecated.
Builder-tool default. Claude Desktop, Cursor, Warp, Raycast, Granola, Superhuman, Arc — every AI-native tool ships Mac-first. Fewer "not supported on Windows yet" walls.
Privacy posture. Secure Enclave + FileVault + per-app permissions give a personal agent sandboxing that Windows lacks without third-party tools.
Community gravity. The indie AI builder community on X is ~80% Mac. Examples, libraries, and Stack Overflow answers all assume Mac.

Don't have a Mac? The Blueprint includes a Windows footnote — reduced features, but functional. The Always-On Path (Part 1) is OS-agnostic and works either way since it runs on Linux.

Beyond v1

What's added in v2.

v1 was generic "build an AI agent." v2 covers the chapters that distinguish "demo" from "production" — the things that kill 80% of attempts.

Multi-agent orchestration

Running 2+ agents that share state. The Mission Control Telegram pattern.

Token + cost observability

Drop in the Agent Fleet Dashboard. Know what your agent costs you, every day.

Identity / SOUL files

How to give an agent a persistent voice that doesn't drift over months.

Memory architecture

File-based vs vector DB. Most "agents" overcomplicate this — when each is the right call.

Telegram control plane

Single mission-control group that scales from one agent to a fleet.

Cron + scheduling

The six-cron pattern: post / reply sweep / outbound / analytics / self-reflect / backup.

Failure modes you'll actually hit

Context overflow, tool-loop infinity, rate limits, model deprecation, identity collapse.

VPS hardening, same day

UFW, SSH keys, Fail2Ban, root disable. Before the agent touches production.

Migration path

When to graduate Mac → VPS as the agent grows up. Failure modes and rollback.

The Blueprint

Both paths. One price.

Part 1 + Part 2 + the v2 chapters. Copy-paste configs for both paths. Lifetime updates.

$97

One payment. No subscription. Instant download.

Get the Blueprint — $97 →

Want the 300-skill library too? Add the AI Skill Library Access at checkout for +$147 — 299 skills, 13 starter packs, 11 done-in-30-days recipes. Order-bump, no second checkout.

Who this is for

And who it isn't.

For you if

You're an operator who wants an agent that ships, not a 200-page textbook
You can run a terminal command and edit a config file
You want to know what your agent actually costs you
You'd rather have one working path than ten frameworks to compare

Not for you if

You want done-for-you (that's DFY at $1,997, not the Blueprint)
You're looking for theory or "What is an AI agent?" — this assumes you know
You won't open a terminal even once
You don't have a Mac AND don't want a $5/mo VPS

Ready to ship?

Both paths. v2 chapters. Copy-paste configs. $97, one payment, lifetime updates.