Skip to Content

Take back control of your AI.

I wrote the Open Source Server Stack series: 9 guides and 60+ production configuration files, so you can build your own personal AI assistant, hosted on your server and under your control.

Why now?

In 2026, your SaaS “AI assistant” is a service you rent, not a tool you own. Four risks nobody spells out when you sign up:

  • The rules change without notice. Yesterday 128k tokens of context. Today 32k. Tomorrow?
  • Your data sits on servers in the USA. Client drafts, strategic decisions, sales proposals — stored with whom?
  • Costs keep climbing. ChatGPT Plus went up 3 times in 18 months. ChatGPT Team at 100 USD/month for 5 people = 1,200 USD/year.
  • No persistent memory. You re-explain the project context 5 times a day. 15 min a day, times 200 working days.

4 pillars of a serious personal AI

Secure VPS

Hardened SSH, WireGuard, Caddy + TLS, encrypted secrets. The foundation that makes everything else possible.

Persistent memory

Cognee + pgvector + autoRecall. Your assistant remembers your projects, decisions and preferences.

Multi-agent

4 agents with separate permissions (web, comm, system, main). Secure by design.

Monitoring

Prometheus + Grafana + Loki. You see what's running, what it costs, what breaks.

The 10 guides of the Protocol

Each guide stands on its own (buy them one by one), or get everything in the Complete Pack.

Guide 0 — Introduction & Prerequisites • FREE

32 pages. The map before the build. Understand the stack, the costs, the technical level required.

Guide 1 — Secure server • CHF 39

9 phases. From a bare VPS to a production server. SSH, WireGuard, firewall, Caddy, fail2ban.

Guide 2 — Encrypted backups • CHF 39

Restic + rclone + retention. Tested backups, restore drill, multi-VPS.

Guide 3 — Secrets & Vault • CHF 39

sops/age, Vaultwarden, rotation. Your secrets encrypted, decrypted in RAM only.

Guide 4 — Team messaging for your assistant • CHF 39

Bot, PAT, threads, bindings, slash commands. Your chat interface with your assistant.

Guide 5 — Persistent memory • CHF 39

Cognee + pgvector + autoRecall. The layer that turns a chatbot into an assistant.

Guide 6 — Multi-agents and approvals • CHF 39

4 agents with separate permissions, Docker sandbox, exec approvals.

Guide 7 — Controlling AI API costs • CHF 39

Smart routing, prompt caching, monitoring. Up to -65% on your AI API bill, depending on usage. Not everything: loaded skills and plugins consume tokens as soon as they initialise.

Guide 8 — Monitoring & Observability • CHF 39

Prometheus + Grafana + Loki. See what's running, in real time.

Guide 9 — Deploying a server with Ansible • CHF 39

Deploy a new VPS in 8 minutes. Inventory, roles, idempotent playbooks.

Packs

Guides Pack

8 guides + 8 appendices

CHF 149

All the written documentation of the 8 main guides. For those who want to understand the stack in depth, without the config files.

See the Guides Pack

Complete Pack

8 guides + 8 appendices + 60+ configuration files

CHF 249

Everything in the Guides Pack + the 60+ config files tested in production. That is what saves 30-40 hours of trial and error.

See the Complete Pack

Want some help?

Free assessment

20-min video call

We look at your situation together (technical level, goals, deadline, budget). I tell you honestly which path gives you the best return.

Book the assessment

Assisted Installation

CHF 597

1-2 video sessions (4-6 h in total). We install the critical foundation together. You keep full ownership. 60 days of support included.

Book Assisted Installation

Turnkey Installation

CHF 2,497

I install it for you. Lead time 7-10 working days. Custom documentation + 1 training session + 60 days of support.

Request Turnkey Installation

What it really costs

USD 7 / month

Infomaniak Lite VPS (4 GB RAM)

~USD 57 / month

Monthly total (VPS + LLM API + extras)

4-8 h

Setup with the Complete Pack (vs 80-150 h from scratch)

Frequently asked questions

The bare minimum (Guides 1 + 3) requires knowing how to open a terminal and copy-paste commands. That's all. For the advanced concepts (Guides 5-7), some familiarity with Docker helps. If that sounds daunting, Assisted Installation (CHF 597) or Turnkey Installation (CHF 2,497) is the right path.

On my current setup: about USD 57/month (7 for the VPS, 46 for the LLM APIs, 4 for extras), for 200-400 requests a day, 1.5M tokens a day, 3 users. Compare that with about USD 72/month in ChatGPT/Claude/Notion AI subscriptions for equivalent use.

Everything is open source (OpenClaw, Cognee, pgvector, Docker, Caddy). You get the guides in PDF + Markdown + 60+ config files. Even without support, your installation keeps running. No dependency on a proprietary service.

With the Complete Pack: 4-8 h for a working system, 20-30 h to master it fully. Without help: 80-150 h. With Assisted Installation: 4-6 h of your time on video calls. With Turnkey Installation: a 1-hour briefing + 7-10 working days for delivery.

Yes. Everything runs on your VPS (with the hosting provider of your choice): conversations, memories and decisions stay 100% with you. For the AI model, you choose according to how sensitive the data is: when confidentiality requires it, a local model running on a machine you control, and nothing leaves; when it is acceptable, an external API (OpenAI, Anthropic, etc.), more powerful, while you control which data goes there through the context you build. Both can live side by side in the same installation.

Yes. Guide 0 is free: 32 pages covering the architecture, the costs and the technical level required. You can decide afterwards, with full knowledge.

Your AI. Your server. Your data.

Start wherever it makes sense for you.

Federico Solutions (Switzerland)