Set up… on the connected server's card (Hosts page), or Set up this host in the host switcher or the command palette, brings a server to what this device has, so its agents work as they do here.
| Step | How | Done when |
|---|---|---|
| System packages | git, curl, unzip, a compiler, with the server's package manager (apt-get, dnf, zypper, pacman, apk; bash too on Alpine, whose agents' installers need it), through sudo (doas where there is no sudo) unless enowX runs as root | git, curl, unzip and make are there |
| mise | curl https://mise.run | sh, its shims put on PATH in ~/.profile (which the daemon's login shell reads), ~/.bashrc, ~/.zshrc |
it is there |
| Toolchains | mise use -g <tool>@<version> for Node, Bun, Deno, pnpm, Python, Go, Rust, Java at this device's versions |
the server has the same to the minor, or a newer one (macOS's own Python isn't worth going back to) |
| Git user | git config --global user.name/email |
they match |
| GitHub CLI | gh through mise, then gh auth login --web: a code to enter in a browser. No token leaves this device. |
gh is there / gh auth status passes |
| Agents | each agent's own installer | its command is there |
Agent sign-in (opt-in, in enowx.toml) |
the agent's sign-in, a code to enter in a browser | — |
| Xpra (when this device has it) | as Install Xpra does (core/display) |
it is there |
Skills, subagents, commands, MCP servers and settings go with Sync environment (sync.md), which the page links to; projects with Copy to another host (transfer.md).
- The app scans this device (
core/setup, in the app) and the server (setup.scan, on its daemon): OS, package manager, the programs above with their versions, git's user, gh's sign-in. Programs are looked for on the daemon's PATH, taken from a login shell, so version managers count. - This device's scan makes the manifest, which Save enowx.toml… writes and Load enowx.toml… reads (to set up servers alike, or keep it in a repo).
- The plan lists each step with what the server has; those it lacks are picked. Show commands is the dry run: every step's shell, before anything runs.
- Set up runs the steps picked as one script in a session of the server, whose terminal shows them and takes what they ask (a sudo password, a sign-in). Each step runs in a subshell that checks before it changes anything, so running again is safe; a failed step doesn't stop the next. Each records its state in
~/.enowx-app/setup/status, which the page reads every 2 seconds to mark it running, done or failed.
Checked by scripts/test-distros.sh, which runs each in a container (docker): the daemon starts, a session's terminal works, the system packages and git steps install what they say, and a second plan has nothing left to do. On Ubuntu 24.04, Debian 12, Fedora 41, Arch Linux (amd64; emulated on arm64 machines, where pacman's download sandbox is turned off for the test), Alpine 3.21 and openSUSE Tumbleweed.
Before installing the daemon, the app readies the server (hostconn/prepare.sh), as root where needed (root, else sudo or doas without a password; it says what to run otherwise): CA certificates where there are none (minimal images and containers), which terminals need to download their library over HTTPS; and on musl systems (Alpine) the glibc compatibility layer (gcompat, libstdc++), since that library is built for glibc.
Connecting to a server scans both ends in the background; its card then says how many steps it lacks ("Set up · 3 to bring"), when this device got a new toolchain or agent since, say.
By default enowX reaches tailnet servers through the tailscale command the device has (core/tailnet). A host's "Through enowX's own tailnet" option instead dials its SSH over a tailnet device of the app's own (transport/owntailnet), so the server is reached without Tailscale on this machine: it signs in once from a browser and keeps its state in ~/.enowx-app/tsnet. ENOWX_TAILNET_CONTROL points it at a self-hosted coordination server (Headscale) instead of Tailscale's. It is part of the app only; the server daemon doesn't carry it.