A founder-grade course that takes you from 'ChatGPT user' to operator of a self-hosted AI agent that runs on your own VPS, talks to you on Telegram, manages your inbox and calendar, and works while you sleep.
Get a free inside look at “Build Self-Hosted Agentic Infrastructure in a Weekend” — the video overview.
By the numbers
You pay $20/month for ChatGPT Plus and $20/month for Claude Pro, and you extract real value. But consumer AI tools have a hard ceiling: they don't remember, don't schedule themselves, don't have access to your stack, and don't show up where you work — they wait behind a tab. So your calendar still has unprepped meetings and you still do the same shallow ops work every single morning. The next move, the one 99% of founders never make, is to stop being a customer of someone else's chatbot.
An AI employee is a software agent you hire (provision a VPS), onboard (configure with the 9-file persona spec), and manage (review output, set guardrails, course-correct). It runs on infrastructure you own, in a contained environment, 24/7, with its own Google Workspace identity, its own Notion login, and its own calendar. The diagnostic is the Four Pillars — Isolation, Persistence, Identity, Channels. Consumer tools score 1-2 of 4; a self-hosted agent scores 4 of 4.
Cost: $24 VPS + $19-25 LLM subscription = about $50/month all-in. Value: 8-15 hours/week of shallow ops reclaimed — morning brief, calendar prep, email triage, task capture. At a conservative $100/hour effective rate, that's $3,200/month of recovered leverage against a $50/month cost. Most founders break even before lunch on day one; the rest is upside.
E¹ Foundation (the stance shift and the founder economic case). E² Build (provision, harden, install, tunnel — plus the 60-minute recovery drill). E³ Connect (Telegram, WebUI, WhatsApp, and the channel-decision rule). E⁴ Configure (the 9-file persona spec across governance, personhood, capability). E⁵ Operate (the four canonical workflows and the 2-week measurement protocol). E⁶ Harden + Capstone (secrets, backups, the 12-point audit, the Lethal Trifecta threat model, and the AI-Employee-as-a-Service play).
6 module decks (~30 slides each), 24 trainer documents (guide, workbook, job aid, facilitator notes per module), 36 brand-locked infographics, 18 video overviews, 6 audio deep-dives, 6 data-table comparisons (stack matrix, VPS providers, channels, 9-file responsibilities, workflow ROI, productization tiers), and enriched briefings synthesized across hundreds of deep-research sources. Lifetime access and free updates as the framework evolves.
Independent research (LSE/Protiviti) puts average AI time savings at 7.5 hours/week and 11 hours/week for trained users — directly validating the course's 8-15 hours/week reclaimed claim.
St. Louis Fed found a third of daily AI users save 4+ hours per week and are ~33% more productive per AI-assisted hour, grounding the economic case in peer-reviewed survey data.
6 modules, each ending in a deliverable.
An AI tool waits behind a login and answers when asked. An AI employee runs 24/7 on infrastructure you own, remembers across sessions, initiates work on a schedule, has its own Workspace and Notion identity, and reaches you on Telegram. The Four Pillars test — Isolation, Persistence, Identity, Channels — distinguishes them: tools score 1-2 of 4, a real employee scores 4 of 4.
No. The build is copy-paste-driven, and the hardening, install, and configuration steps each come with a verification check so you know they worked. The course is explicitly designed for non-developer founders, most of whom complete the first deployment in one weekend.
About $25-50 all-in: a $5-24 VPS plus a $19-25 subscription LLM brain. The course insists on subscription billing over metered API keys precisely to keep a 24/7 agent's cost fixed and predictable.
For the worked example, Hetzner CX22 (~$4.44/mo) or Digital Ocean Basic 4GB ($24/mo) on Ubuntu LTS, paired with a subscription brain — Kimi, GPT Plus, or Claude Pro. Because brain and orchestrator are decoupled, you can swap brains in about five minutes if your needs change.
One focused weekend for the build. The 2-week measurement protocol then takes you from a baseline to 8-12 hours/week reclaimed, with full ownership of all four canonical workflows by week four or five.
A 7am morning brief (priorities, decisions, blockers to Telegram), task ingestion every two minutes (nothing leaks from chat/email into Notion), calendar pre-meeting summaries at T-15 with conflict detection, and email triage every 15 minutes that categorizes and drafts replies — but never auto-sends.
Three-tier secrets management (vault to .env to process, rotated quarterly), automated daily and weekly off-site backups, a 12-point quarterly audit, and architecture against the Lethal Trifecta. The draft-only rule, gateway security, and OAuth scope minimization each break one of the three exfiltration legs.
It's the security model (Simon Willison, June 2025, now adopted by OWASP) describing the dangerous combination of private-data access, exposure to untrusted content, and the ability to communicate externally. An agent with all three can be tricked by a single injected prompt into leaking data. The fix is structural — remove one leg by design — which is exactly what the course's operating rules do.
No, by design. The agent drafts replies and prepares actions, but the founder approves and sends. This is both an etiquette rule and a security control: keeping the agent draft-only removes the exfiltration leg of the Lethal Trifecta.
Nine configuration files — AGENTS, BOOTSTRAP, IDENTITY, SOUL, USER, MEMORY, SKILLS, TOOLS, HEARTBEAT — that turn a generic LLM into a specific employee. They cluster into governance, personhood, and capability, and the whole directory is version-controlled so every behavior change is a commit and rollback takes seconds.
Yes. Once your base agent is solid, the base+fork pattern lets you stand up a specialized variant — a sales agent, a support agent — in about 60 minutes by overriding only the files that differ (usually IDENTITY, SKILLS, and part of TOOLS).
Module 6 covers AI-Employee-as-a-Service: roughly $2-5K setup plus $500-1.5K/month retainer per client, around 95% margin on the retainer. With a proven onboarding playbook, 10-15 clients per operator is about $144K/year. The client's own ROI math (8 hrs/week saved = $3,200/month value) sells the retainer.
You re-apply the six phases. The course teaches durable concerns — foundation, build, connect, configure, operate, harden — not a single tool. The configuration changes when you switch stacks; the architecture and practice carry over. Lifetime access includes free updates as the framework evolves.
Yes. The course assumes zero VPS or terminal experience. Every command is copy-paste-ready, and the 30-minute hardening drill walks you through the entire lockdown step by step. Most non-developer founders complete the build in one focused weekend.
Founding-member pricing, one-time, lifetime access. Deploy a hardened, productized AI employee in a weekend.
