Orchestrate AI coding agents in parallel
AIOAT is a Windows desktop app that runs Claude Code, Codex, Gemini and other agent CLIs side by side — each in its own git worktree, under its own identity — and gives you one place to watch, steer, review, and merge their work.
Install
- Download the installer —
AIOAT_x64_en-US.msi(recommended) orAIOAT_x64-setup.exe. - Run it. AIOAT installs per-machine like any Windows app; upgrades install straight over the old version and keep your data.
- Start AIOAT. The welcome screen offers to install Claude Code (or another agent CLI) if none is found — you can also do that later under Settings → Agents.
Requirements: Windows 10/11 with WebView2 (preinstalled on Windows 11), git on PATH for the git features, and at least one agent CLI you're logged in to. Optional: gh for GitHub issues/PRs, Node for npm-installed agents, OpenSSH (ships with Windows) for remote servers.
AIOAT updates itself: when a new version is out, the version in the status bar grows an amber chip — click it to download and install, and the patch notes appear on the next launch.
The first five minutes
- Pick your identity. Open Profiles (bottom of the left rail). AIOAT ships with Personal and Work — each holds its own git name/email, SSH key, GitHub token, and agent API keys, stored in Windows Credential Manager. Agents only ever see the profile they run under. More →
- Register a repo. Code → Repositories → + Add Repository. Pick any folder — if it isn't a git repo yet, AIOAT offers to
git initit. - Launch an agent. Agents → + New Agent. Choose the CLI, the repo, and keep isolated worktree on — the agent gets its own branch and folder, so two agents never trample each other. No repo? Choose No repo (scratch). Prefer chatting to a terminal? Agents → Chat. More →
- Work in parallel. Launch a second agent on the same repo. Watch both under Agents → Sessions; anything that stops to ask you a question surfaces in Approvals.
- Review and merge. When an agent is done, its branch appears in Code → Review inbox: diff, test run, verification gate, merge queue. More →
Finding your way around
The left rail has six destinations; each opens its section in the sidebar. Collapse the rail to icons with the « button at the bottom.
| Section | What lives there |
|---|---|
| Home | Dashboard: live status, run history (searchable, including transcripts), costs, tech-debt ledger. |
| Agents | Sessions (every live agent with state, context %, cost), Chat (a proper chat with Claude instead of a terminal), and Approvals (everything waiting on you). |
| Code | Repositories, Source control (stage/commit under the bound identity), Review inbox, Checkpoints. |
| Orchestrate | Foreman (delegate a goal to a coordinator agent), GSD planning, Canvas overview. |
| Knowledge | Context graph (shared knowledge agents start from) and the Importer (CLAUDE.md, subagents, …). |
| Servers | Remote hosts over SSH — run agents on your own server from AIOAT. |
Keyboard essentials
| Keys | Action |
|---|---|
| Ctrl+Shift+P | Command palette — every action in the app, searchable. |
| Ctrl+, | Settings. |
| Ctrl+F | Search the focused terminal's scrollback. |
| Ctrl+V | Paste into a terminal — pasted screenshots are saved and their path typed for the agent. |
| Ctrl+Alt+A | Global show/hide (configurable in Settings). |
Panels are dockable: drag a tab to a pane's edge to split, to its center to group, or float it via the palette. Terminals live in a detached session host, so closing the window does not kill your agents — reopen AIOAT and the sessions reattach.