In one line
My R&D lab – where I try new AI tooling before I bring it to work.
Context & problem
I wanted a state-of-the-art setup to turn ideas into running programs, and a growing token bill for my personal agents.
What I did
- Three-device setup over Tailscale: Mac Mini M4 as always-on server, gaming PC (RX 7900 XTX, 24 GB) for local LLM inference, MacBook on the road
- Moved the LLM power behind my agents onto a local model, shared across all my automations
- Personal agent controlled via Telegram (migrating from OpenClaw to Hermes); cheap or free OpenRouter models with spending limits for harder tasks
- Project template idea: ideas brainstormed on the train flow straight into a repo and the next steps
Result / impact
A greenfield to experiment freely, and local models save me money on paid API tokens.
Details
- Safety & data protection: self-hosted, data stays at home.