If I don’t have agentic LLM setup in this particular way, I end up spending more energy fixing the output than I save on development effort. This setup is built upon a foundational preference for executing a few ambitious/expensive prompts a week instead of the more common “conversational” approach to agentic AI. That shit frustrates me.
This repository has two purposes: (1) Online backup of my agentic LLM configuration. (2) Streamline deployment of updates to this configuration across my many installation instances.
The scripts provided by this repository handle the following (idempotently):
~/.claude/) with the files in claude-code/Note: Install scripts overwrite
CLAUDE.mdand a few user settings.
curl -fsSL https://claude.toasterparty.net/install.sh | bash
irm https://claude.toasterparty.net/install.ps1 | iex
Runs install.sh, then serves a directory to claude.ai/code and the Claude app through a systemd user service that survives reboots. Every session started from the cloud runs in that directory. Repeat with another --dir to serve more repos. Logging in takes a terminal the first time.
curl -fsSL https://claude.toasterparty.net/start-server.sh | bash -s -- --dir ~/git/repo
--uninstall removes the service for one --dir, or every one when --dir is omitted. Login, workspace trust, and linger stay in place.
curl -fsSL https://claude.toasterparty.net/start-server.sh | bash -s -- --uninstall [--dir ~/git/repo]
Set CLAUDE_DIR to deploy into a directory other than the default ~/.claude/.
Linux/MacOS:
curl -fsSL https://claude.toasterparty.net/install.sh | CLAUDE_DIR="$HOME/.claude-work" bash
Windows:
$env:CLAUDE_DIR = "$HOME\.claude-work"; irm https://claude.toasterparty.net/install.ps1 | iex
$env:CLAUDE_DIR = "$HOME\.claude-personal"; irm https://claude.toasterparty.net/install.ps1 | iex