Windows Terminal grows a native AI agent pane

What it does

Intelligent Terminal is an experimental fork of Windows Terminal that bakes agent support directly into the console. It docks an AI agent pane alongside your shells, feeding it live shell output and error context over stdio using the Agent Client Protocol. You can chat, debug failed commands, or delegate background tasks without leaving the terminal window.

The interesting bit

Instead of treating AI as a browser sidebar, this makes it a first-class terminal citizen: it auto-detects ACP-compatible CLI tools like Copilot, Claude, Codex, or Gemini, and spawns long-running agent work in new tabs so your active shell stays focused. The error-detection hook is the quietly useful part—when a command fails, the agent pane can open with the stack trace already loaded.

Key highlights

• Agent-agnostic: works with any ACP-compatible agent CLI, not just GitHub Copilot, and auto-detects several on first launch.
• Context-aware by default: the agent pane reads shell output across tabs and panes, so you don’t have to copy-paste errors for help.
• Error detection lights up when commands fail, letting you send context straight to the agent for diagnosis or suggested fixes.
• Runs as a local transport layer: it does not call cloud APIs itself and discards session data from memory when the session ends.
• Requires Windows 11 22H2 or later; still explicitly experimental.

Caveats

• Windows 11 22H2 or later only; Windows 10 users are out of luck.
• Requires a supported agent CLI and subscription—data routing and privacy terms depend entirely on whichever agent vendor you choose.
• Microsoft collects usage telemetry by default, though the README notes you can turn it off.

Verdict

Give it a spin if you live in Windows Terminal and want AI assistance without juggling browser tabs. Skip it if you’re on Windows 10 or prefer your terminal to stay strictly text-in, text-out.