Case study · Members only Agents
Our own AI assistant with Claude Code and Obsidian: a real-world case
How the assistant we use daily at GamzyHost is built: tasks, one Claude Code chat per project, approvals for sensitive actions and memory in Obsidian.
Tested on October 16, 2026 in Claude Code 2.1.283 · Windows Server 2022. System in daily use at GamzyHost. How we test
Our own AI assistant is a Node program that runs 24/7 on a VPS with Windows Server 2022. It takes tasks by voice, through the web or through Discord, and hands them out to Claude Code sessions, one per project. Those sessions work with MCP servers and SSH. Anything sensitive goes through our approval, and whatever it learns ends up in an Obsidian vault organized by project.
Below you'll find the full architecture and the reasoning behind it, the table of what it does on its own and what it asks us about, how its memory gets tidied every night, three things that went wrong, what it costs, and how to build a minimal version in a weekend.
What the full version includes
- The full architecture of our assistant, with a diagram of every piece and why we chose it
- The real table of what it does on its own, what it asks us first and what it never does without approval
- How its memory is organised in Obsidian: project cards, per-part notes and the nightly routine
- Three mistakes that cost us and how we fixed them, with numbers
- What it costs and what we would not recommend copying
- A plan to build a minimal version in a weekend
The full version is for members
Articles with our own method, the real configuration and the full code are for readers who support the blog with a membership: €4.99 every 3 months. Everything else stays open to everyone, and so does the AI news board.
You are signed in, but your account has no active membership.
Already a member? Sign in
