onstr on Nostr: Run it locally. Get a mac mini or something with ~32gb of ram. With that you can run ...

Run it locally. Get a mac mini or something with ~32gb of ram. With that you can run really good open models like Qwen3.8-27b.

Use Hermes for the harness, and create an Obsidian vault for the agent brain and memory. Connect with Tailscale for remote access from your other devices.

Maximum privacy (nothing leaves your device), and no subscription. Performance of these open models is so good now I think this is the way.