Skip to content

wdx documentation

wdx runs AI coding agents (Claude Code, Codex, Cursor and Antigravity) on your machine, each in a terminal that outlives the window you started it from, and often in a git worktree of its own. A daemon keeps them running; the wdx CLI and TIDE, its Terminal IDE, are how you and your scripts reach them.

  • What is wdx: what it does and what it is not.
  • Install: one command on Linux or macOS.
  • Quickstart: a project, an agent and TIDE in five minutes.
  • Concepts: the words the rest of the docs use.

One per feature, each with a session to try, how it shows in TIDE, how it behaves and its commands:

  • Projects: register, clone, close and reopen repositories.
  • Terminals: shells that outlive the window and the daemon.
  • Agents: AI CLIs wdx runs, reads and resumes.
  • Child agents: agents that start agents of their own.
  • Messages and sending: tasks for agents from scripts and other agents.
  • Worktrees and landing: a branch per agent, a diff, and landing with no merge commit.
  • Docs: designs kept in wdx with a revision per change.
  • Plans and Running plans: a feature as tasks for agents, run and landed by wdx.
  • Hub and Running a hub: reach your machines from each other, and host your own.

The Terminal IDE, page by page, with its screens live: Layout and keys, Agents, Worktrees and diffs, Code, Docs and plans, The menu, Settings and Stats.

The AI CLIs wdx runs agents on, side by side, and a page each: Claude Code, Codex, Cursor and Antigravity.

How wdx works inside, each page around a diagram: the overview, Daemon and service, Holders and PTYs, State in one SQLite file, The API, Reading an agent’s state, Worktrees and watchers and The hub.

These pages are generated from wdx itself, so they say what wdx <command> --help says.

The Roadmap, the FAQ, Contributing and the License.