The tools I keep running.

A file-first memory system for AI agents. Markdown files you own and version, the server is only an adapter.

The best local LLM setup for your machine: picks llama.cpp or MLX, installs it, fetches and sizes the model, serves one OpenAI-compatible URL.

Benchmarks OpenAI-compatible endpoints the way agents use them: decode and prefill speed, tool-calling accuracy, needle retrieval, tagged runs.

Keep it alive, keep it current: self-update from a release manifest with rollback, and a keep-alive daemon under launchd, systemd or nothing.

Terminal and coding agents in a browser tab. PTY sessions survive the tab closing, scrollback replays on reconnect.

Local-first keystroke and activity tracking for macOS. Searchable, charted, never leaves the machine.

Phone backup to a USB drive, simple enough for your parents. Offline, resumable, verified.