AI agents

Based installs no coding agent for you. Developer tools covers picking one from the menu. What Based does is keep every agent you install supplied with the same MCP servers and skills. One program, agentctl, is the only thing that writes them, so a server you add shows up in Claude Code, Codex, OpenCode and the rest at once.

This is on every Based edition except the pure ones.

Where servers come from

source	how its servers get to your agents
an app you install	the app's MCP server comes in its own package, and leaves with the app. LibreOffice, Thunderbird, Chromium, GIMP, Inkscape, Blender, FreeCAD and Audacity have one today
your organization	signing in to a Work or School Account adds the servers it publishes, each with a switch on the sign-in so you can leave some out. Accounts has the sign-in
you	AI Tools' Add, or `agentctl add`

A new server is on in every agent, except one that runs code on your machine, which starts off until you switch it on. Thunderbird's send and delete tools ask you first, in the agents that can ask.

AI Tools

Open Setup, then AI Tools in the Omarchy menu, or AI Tools from your apps. It has three pages:

Servers, grouped by where each came from, with a switch per agent. Your organization's servers have Refresh, which brings them back in step with what you approved, and Leave, which takes them out of every agent and keeps your sign-in.
Skills, the Agent Skills folders, with a switch per agent. Add a folder that holds a SKILL.md.
Agents, the ones installed here and whether agentctl manages each.

From a terminal

agentctl status                          # every server and agent, and its state
agentctl add <name> --command <cmd>      # a server that runs a program
agentctl add <name> --url <url>          # a server over HTTP
agentctl disable <name> --harness codex  # off in one agent
agentctl skills add --from <dir>         # a skill folder
agentctl doctor                          # what keeps a server from reaching an agent

Each command that changes something also updates every agent before it returns.

What it changes in an agent's files

Only the MCP section, such as Claude Code's mcpServers or Codex's [mcp_servers], and the agent's skills folder. Your logins and other settings stay as they are. A token you give it goes into your keyring, and the agent's file names only agentctl run <name>, so no agent's file holds a secret.

A server you add inside an agent directly is taken into agentctl's list, not dropped. A server you switch off by deleting it in one agent stays off there. An edit to a server agentctl defines doesn't last, though: the next sync puts agentctl's version back. Change it with agentctl add <name> --replace.

On a Mac

The Omarchy install on macOS doesn't include agentctl. It's one public download into the ~/.local/bin that install already puts on your PATH, where v is agentctl's newest release:

v=0.4.1; a=$(uname -m | sed s/x86_64/amd64/); curl -fsSLo ~/.local/bin/agentctl "https://downloads.basedlinux.com/agentctl-binaries/$v/agentctl_${v}_darwin_$a" && chmod +x ~/.local/bin/agentctl

Then agentctl add and agentctl skills add work as above. Nothing syncs by itself on a Mac, so run agentctl sync after installing a new agent. Your organization's servers don't reach a Mac.