Federated_Computer/coredesktop
3
0
Fork 0
Code Issues 1 Pull Requests Actions Packages Projects Releases Wiki Activity

Add the configure app

Browse Source
This commit is contained in:
david
2026-10-26 15:16:53 -06:00
parent 63d56da089
commit bbd6b83f64
21 changed files with 1010 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+38
View File
@@ -202,6 +202,44 @@ Use "sibling list" to see what's installed and "sibling remove {name}" to uninst
To make your own repository work as a sibling project, see [skills/sibling-project-format.md](skills/sibling-project-format.md).
## Configure
Some of what CoreDesktop installs needs you to sign in or make a choice before you can use it. To see what's set up and set up the rest, run in the terminal:
```
configure
```
It covers:
- **Your Federated Computer Core.** Give it your Superpanel's address (e.g. https://superpanel.example.com), your Core email, name and password, and it adds Superpanel to your applications, joins this computer to your Core's Tailscale network, adds your Nextcloud (calendar, contacts and files) and email to GNOME's Online Accounts, and mounts your Nextcloud files at /mnt/example.com ("example.com" in the Files sidebar). For Nextcloud and the files it creates a Nextcloud app password, which you can revoke in Nextcloud under Settings > Security.
- Signing in to Claude Code, opencode and Hermes Agent and choosing their model providers.
To set up one thing, name it, e.g. `configure hermes`; `configure --list` shows the names.
## Sandboxed AI agents
CoreDesktop installs the AI coding agents opencode, Hermes Agent and Claude Code, and runs them in a sandbox. In the sandbox, an agent:
- can change files only in the directory you start it from (plus its own settings and a few package caches), so start it from a project directory
- can't read your SSH keys, password stores, browser profiles, cloud and Git credentials, or the other agents' logins
- can reach only AI providers and common developer sites: GitHub, Gitea, npm, PyPI, crates.io, RubyGems, the Go module proxy and archlinux.org
Typing `opencode`, `hermes` or `hermes-agent` runs it in the sandbox. To run one without it, type `unsandboxed opencode`. Hermes' gateway, the background service that lets you talk to Hermes through Matrix and other messaging apps, runs in the sandbox too, and can reach the Matrix server in Hermes' settings. For other messaging services, add their domains as shown below. Claude Code uses its own built-in sandbox for every command it runs (see `/sandbox` inside Claude).
Before using an agent the first time, sign in and choose a model provider with `configure` (see below).
If an agent needs a site or folder the sandbox blocks, add it in `~/.config/coredesktop/sandbox/base.json` (for every agent) or `~/.config/coredesktop/sandbox/opencode.json` (for one agent):
```json
{
"network": { "allowedDomains": ["docs.example.com"] },
"filesystem": { "allowWrite": ["~/notes"] }
}
```
The sandbox is Anthropic's [sandbox runtime](https://github.com/anthropics/sandbox-runtime), which uses bubblewrap and socat. The default rules are in [sandbox/config/](sandbox/config/).
## mDNS Networking
If you want to use mDNS networking on your LAN (the ability to log in to other local machines using the format "machine.local"), then you should replace /etc/nsswitch.conf with the one in this repository and add /etc/mdns.allow. These two files are formatted to avoid any conflicts with systemd-resolved that may cause nss-mdns not to work. See: https://wiki.archlinux.org/title/Avahi#Hostname_resolution (section 4.2) for more information.