# Intelligent Terminal: Microsoft's Windows Terminal fork with an ACP agent pane

> Intelligent Terminal is an experimental fork of Windows Terminal that puts an Agent Client Protocol agent beside your shell. It is Windows-only, experimental, and it asks you to install an agent CLI before it can do anything.

**microsoft/intelligent-terminal** — A fork of Windows Terminal with native agent integration, right in your command line.

- Repository: https://github.com/microsoft/intelligent-terminal
- Website: https://devblogs.microsoft.com/commandline/announcing-intelligent-terminal-version-0-1/
- Stars: 2,035 · Forks: 164
- Language: C++
- License: MIT
- Published: 2026-09-10 · Updated: 2026-09-10 · Language: en
- Canonical page: https://hysenlabs.com/projects/microsoft-intelligent-terminal

## The problem Intelligent Terminal solves: shell output that an agent can read

A normal terminal emulator and a coding agent live in separate windows. You run a command, it fails, you select the error text, switch windows, paste it into the agent, and describe what you were doing. Intelligent Terminal removes that loop by forking Windows Terminal and adding an agent pane inside the same window. The README states that "the agent has context on your shell output, no copy-pasting needed." The intended user is a Windows developer who already runs an agent CLI and wants it adjacent to the shell rather than in a browser tab or a second application. It is not a new terminal engine. Tabs, profiles, themes, settings, shells and keyboard shortcuts are inherited from Windows Terminal and, per the README, "all work the way you expect." The agent layer sits on top of that. The project is built around the Agent Client Protocol, so it is not tied to one vendor: the README says it "works with any Agent Client Protocol (ACP)-compatible agent CLI." GitHub Copilot CLI is the default, but the auto-detection list also names Claude, Codex, Gemini and OpenCode.

## How the agent pane, error detection and session hooks fit together

The mechanism is a terminal that watches its own shell integration and feeds selected context to a separate agent process. On first launch you pick an agent; the terminal scans the machine for ACP-compatible CLIs and, if it finds none, defaults to GitHub Copilot CLI and installs it through WinGet. From then on the agent pane is a panel inside the window, placed top, bottom, left or right, toggled with Ctrl+Shift+. and focused with Ctrl+Shift+I.

Two settings drive the automatic part. Error detection lets the terminal notice when a command fails. Error suggestions then sends those detected errors to the agent for fix suggestions. Ctrl+Alt+. opens the agent pane with error context already attached, which is the shortcut that makes the feature worth having. The agent does not act on its own: the README is explicit that "the agent pane does not run commands in your shell without your explicit approval," and you choose to run a suggested command, copy it, or dismiss it.

A third layer tracks sessions. With agent session tracking (hooks) enabled, the terminal tracks active agent sessions and their status in the session management UI. Turning off Settings > Agents > Sessions does not remove hooks that are already installed; the setting controls their installation and updates, and the README notes existing hooks "may continue to report activity." That is a real asymmetry worth knowing before you toggle it off expecting a clean stop.

Configuration is per pane and per profile. Each agent pane can override its model at runtime with /model, and changing the global model setting overrides every pane. A profile can also be pinned to a specific agent; for a WSL profile the picker lists agents installed inside that distribution, so an Ubuntu profile can run a Linux-side agent while unconfigured profiles keep the global one.

## Installing Intelligent Terminal with WinGet and getting the agent pane running

The README lists three install paths: the Microsoft Store (recommended, for automatic upgrades), WinGet, and a direct App Installer download for x64, arm64 and x86. The prerequisite is Windows 10 2004 (19041) or later, plus a supported agent CLI and subscription. The WinGet command installs the package by its exact identifier:

```powershell
winget install --id Microsoft.IntelligentTerminal -e
```

After installation, launch the terminal. The first screen asks you to choose an agent. If an ACP-compatible CLI is already present, it appears in the list; if not, the terminal falls back to GitHub Copilot CLI and installs it via WinGet. If you are not authenticated, the agent pane walks you through sign-in. For GitHub Copilot Enterprise, press E at the sign-in prompt and enter your enterprise host, for example your-org.ghe.com. The README says the last host you used is remembered.

One failure is common enough that the README calls it out: a PowerShell execution policy that blocks the profile. The symptom is "running scripts is disabled on this system" or an UnauthorizedAccess error, and the terminal cannot initialize shell integration without the profile. The documented fix is:

```powershell
Set-ExecutionPolicy -Scope CurrentUser -ExecutionPolicy RemoteSigned
```

Once the pane is open, the first useful test is to run a command that fails and press Ctrl+Alt+. The pane should open with the error as context. If you want a provider other than the default, add one under Agent settings with a Base URL, Model ID and optional API key, then select it from /model. API keys go to Windows Credential Manager, and keyless local endpoints such as Ollama remain supported.

## Windows-only, experimental, and dependent on an agent you install separately

The first limitation is platform. This is a fork of Windows Terminal, the install paths are the Microsoft Store, WinGet and an App Installer package, and the prerequisite is a Windows build number. There is no Linux or macOS install described anywhere in the README. If your team develops on macOS or Linux, this project is not a candidate regardless of how appealing the agent pane sounds.

The second is that the terminal is not the intelligence. It ships no model. Every useful behaviour depends on an ACP-compatible agent CLI that you install and pay for, and the default path installs GitHub Copilot CLI through WinGet when nothing is detected. A machine with no agent, no subscription, or a locked-down WinGet is left with a Windows Terminal fork and an empty pane.

The third is maturity. The README describes the project as "an experimental fork of Windows Terminal." The release history is consistent with that: v0.1.18 on 2026-07-10, v0.2 on 2026-08-10, and v0.2.23 on 2026-08-28. The last push to the repository was on 2026-09-10, so work is ongoing, but the version numbers are still in the 0.2 range. Anyone treating this as a drop-in replacement for a terminal they depend on for production work should weigh that.

There is also an opt-out trap. Disabling Settings > Agents > Sessions hides the session UI but does not uninstall existing hooks, which the README says may keep reporting activity. If your reason for turning it off is privacy or noise, the setting alone may not achieve what you want.

## Intelligent Terminal compared with Windows Terminal plus a separate agent CLI

The honest alternative is the setup most people already have: stock Windows Terminal, or another emulator, with an agent CLI running in its own tab or window. The difference is context plumbing. In the separate-window setup, the agent only knows what you paste. In Intelligent Terminal, error detection and error suggestions push detected failures into the agent automatically, and Ctrl+Alt+. opens the pane with that context attached. The pane also has access to shell output generally, which is the whole point of the fork.

The trade-off runs the other way too. Stock Windows Terminal has no agent code in it, so it has no agent-related settings, no hooks to manage, and no dependency on a subscription. It also receives Microsoft's own updates on its own schedule. Choosing Intelligent Terminal means accepting a fork that tracks upstream Windows Terminal while adding a layer that upstream does not have, and the README's own framing of the project as experimental is the clearest statement of where that leaves you.

Within the agent layer there is a second choice. You can stay on the default GitHub Copilot CLI, or point the terminal at a different ACP-compatible CLI. The README also documents bring-your-own-key providers through GitHub Copilot or OpenCode using an OpenAI-compatible Chat Completions endpoint, with keyless local endpoints such as Ollama supported. If your reason for wanting this terminal is to keep prompts on your own hardware, that path exists, but it still requires an agent CLI that speaks ACP.

## Maintenance, licensing and what the MIT licence does not cover

The repository is not archived, and the last push was on 2026-09-10, so the project is receiving commits. Releases have been frequent through the 0.2 line. The practical upgrade path is the Microsoft Store build, which the README recommends specifically because it keeps you on the latest version with automatic upgrades. WinGet installs are updated through WinGet, and the App Installer download is updated by reinstalling from the releases page. The README does not document a rollback procedure, so if a new build breaks your workflow, the recovery path is not described there.

Licensing is MIT, and the repository carries a LICENSE file plus NOTICE.md. The MIT licence governs the code in this repository. It does not grant you access to any agent, model or subscription the terminal connects to, and those are governed by their own terms. The README points to GitHub Copilot CLI as the default agent and links to its own page. Nothing in the README suggests the terminal bundles model access. If you are evaluating this for a company, the question to route to whoever handles procurement is not the terminal's licence but the agent subscription and where prompts are processed, particularly if you enable error suggestions, which sends detected errors to the agent. The README has a Data & Privacy section, and PRIVACY.md and TelemetryEvents.md exist at the repository root, which is where those answers live.

## Conclusion

Adopt Intelligent Terminal if you work on Windows 10 2004 (19041) or later, already pay for an agent CLI such as GitHub Copilot, and want the agent to see your shell output without copy-pasting. Do not adopt it if you need Linux or macOS, if you cannot install an ACP-compatible agent CLI, or if you want a stable terminal: the README calls it experimental, and the newest release is v0.2.23 from 2026-08-28. Before committing, verify three things in this order: that an agent CLI is on the machine or that the WinGet fallback can install Copilot CLI, that PowerShell's execution policy lets the profile run, and that your agent subscription covers the model you intend to select with /model.

## FAQ

### What is Intelligent Terminal?

It is an experimental fork of Windows Terminal with native agent integration, so an ACP-compatible agent CLI runs in a pane inside the same window as your shell. Tabs, profiles, themes, settings, shells and keyboard shortcuts are inherited from Windows Terminal and work as before.

### How do I use Intelligent Terminal?

Install it from the Microsoft Store, WinGet or the App Installer download, then choose an agent on first launch. Open the agent pane with Ctrl+Shift+. and start asking questions; the README notes the agent has context on your shell output, so no copy-pasting is needed.

### What is Microsoft Intelligent Terminal?

It is Microsoft's experimental fork of Windows Terminal with native agent integration, published at github.com/microsoft/intelligent-terminal under the MIT licence. It works with any Agent Client Protocol (ACP)-compatible agent CLI, with GitHub Copilot CLI as the default.

### What is an intelligent terminal in a computer?

In this project the term names a terminal emulator with agent integration built in: Intelligent Terminal auto-detects ACP-compatible agent CLIs, tracks their sessions, and detects command failures so the agent can suggest fixes. The README describes it as an experimental fork of Windows Terminal.

## Sources

- [License: MIT](https://github.com/microsoft/intelligent-terminal/blob/main/LICENSE)
- [microsoft/intelligent-terminal on GitHub](https://github.com/microsoft/intelligent-terminal)
- [Project website](https://devblogs.microsoft.com/commandline/announcing-intelligent-terminal-version-0-1/)
- [README](https://github.com/microsoft/intelligent-terminal/blob/main/README.md)
- [Releases](https://github.com/microsoft/intelligent-terminal/releases)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/microsoft-intelligent-terminal
