Intelligent Terminal
An experimental fork of Windows Terminal with native agent integration right in your command line.
Evidence shows: README clearly states the terminal is a local transport layer, does not call cloud APIs, does not persist conversation history, data is held in memory only and discarded when session ends; provides privacy statement and telemetry disable instructions; user can configure error detection and auto-suggestions, and explicitly states it always asks before running commands in the shell. Deductions: no detailed least-privilege model or sandboxing; no specific sensitive data handling measures (e.g., encryption); dependency security only mentions WinGet, no audit or vulnerability scanning evidence; external effects (e.g., auto-installing Copilot CLI) lack rollback or uninstall instructions; rollback mechanism not mentioned.
Evidence shows: README describes features consistent with settings, such as error detection and command palette; provides troubleshooting tips (e.g., execution policy) and dependency installation doc link. Deductions: dependency availability only mentions WinGet and Store, no offline installation or mirrors; failure messages only mention reporting errors, no specific error handling mechanisms.
Evidence shows: clear target audience (Windows Terminal users), supports multiple ACP-compatible agents, multiple installation methods (Store, WinGet, download); capability boundaries clear (agent pane, error detection, command palette); triggers clear (keyboard shortcuts, slash commands); environment requirements clear (Windows 10 2004+). Deductions: no detailed compatibility differences among agents or custom agent configuration details.
Evidence shows: README structure clear with TOC, installation, configuration, features, privacy; provides installation notes and FAQ link; naming consistent (Intelligent Terminal); provides example screenshots and FAQ; explicitly states experimental stage and known limitations (e.g., subscription required); MIT license clear; versioning via releases and Store updates, but no CHANGELOG; maintenance responsibility clear (Microsoft official). Deductions: version number not explicit in README, changelog not provided.
Evidence shows: high output usability with clear screenshots and feature descriptions; clear marginal value (native agent integration, reduces copy-paste); cost-benefit not detailed but free open source and Store installation. Deductions: no performance benchmarks or user feedback evidence.
Evidence shows: README claims align with code structure (e.g., ACP, error detection); provides privacy statement and FAQ links; but no test results or independent verification. Deductions: claims are descriptive, lack verifiable test evidence; facts and inferences not clearly separated.
- This repository is experimental and may contain undiscovered defects or security issues.
- Auto-installing GitHub Copilot CLI introduces external dependencies; users should understand its data routing and privacy policy.
- Data handling by third-party agent CLIs is not controlled by the terminal; users must review their policies.
- Static review cannot verify actual runtime behavior; testing in an isolated environment is recommended.
What does this agent do, and when should you use it?
Intelligent Terminal is an experimental fork of Windows Terminal that natively integrates Agent Client Protocol (ACP)-compatible agents. It retains all core Windows Terminal features like tabs, profiles, themes, settings, shells, and keyboard shortcuts, and adds an agent status bar, agent pane, agent management, error detection, and command palette. It works with any ACP-compatible agent CLI (e.g., GitHub Copilot, Claude, Codex, Gemini, OpenCode) and auto-detects installed agents. Data is passed locally over stdio, no cloud APIs are called by the terminal, and conversation history is not persisted. Install via Microsoft Store, WinGet, or direct download. Requires Windows 10 2004 (19041) or later and a supported agent CLI (default GitHub Copilot).
Intelligent Terminal acts as a local transport layer, forwarding user prompts, shell output context (recent command output), and basic environment metadata (shell type, OS version) to the selected agent CLI (e.g., GitHub Copilot CLI) via stdio/ACP. It features an agent status bar (bottom, with toggle and error detection icons), an agent pane (dockable, context-aware, with focus switching), agent management (view active agents and past sessions), error detection (automatically detects command failures and suggests fixes), and a command palette (launch agent tasks with ?). All agent features are configurable under Settings > Agent, including pane placement and error detection toggles. The agent pane supports slash commands like /fix, /clear, /new, and more.
- Developers who want inline command help without leaving the terminal can ask questions in the agent pane and get context-aware answers.
- Users encountering command errors can leverage error detection to automatically detect failures and receive fix suggestions. they can click the error icon or press Ctrl+Alt+. to open the agent pane with context loaded.
- Developers working across multiple shells (PowerShell, WSL) can use the agent pane to maintain context across shells.
- Power users can launch background agent tasks from the command palette by typing '?' plus a prompt (Alt+Shift+/).
- Users managing multiple agent sessions can use agent management (Ctrl+Shift+/) to view status and resume past sessions.
What are this agent's strengths and limitations?
- Native integration into Windows Terminal, preserving all existing features and workflows.
- Works with any ACP-compatible agent CLI, offering choice (Copilot, Claude, Codex, Gemini, OpenCode).
- Local transport layer: does not call cloud APIs directly, and does not persist conversation history, enhancing privacy.
- Error detection automatically catches command failures and provides context-aware fix suggestions, boosting productivity.
- Only supports Windows 10 2004 or later; not cross-platform.
- Requires external agent CLI and subscription (default GitHub Copilot requires a paid plan).
- Experimental stage; may have stability issues, and agent features require configuration.
How do you install or deploy this agent?
- Install from Microsoft Store (recommended): visit https://apps.microsoft.com/detail/9NMQC2SSJX24, or use WiGet:
winget install --id Microsoft.IntelligentTerminal -e. 2. Ensure Windows 10 2004 (19041) or later, and have an ACP-compatible agent CLI installed (e.g., GitHub Copilot CLI). 3. If PowerShell blocks scripts, runSet-ExecutionPolicy -Scope CurrentUser -ExecutionPolicy RemoteSigned. 4. Alternatively, download the App Installer from GitHub Releases.
How do you use this agent?
On first launch, choose your agent (auto-detects or defaults to GitHub Copilot CLI). If not authenticated, the agent pane walks you through sign-in. Use Ctrl+Shift+. to toggle the agent pane, Ctrl+Shift+I to switch focus, Ctrl+Alt+. to open with error context. Type /help in the agent pane for commands. In the Command Palette, type '?' then your prompt to start a task. Configure agent settings in Settings > Agent.
How does this agent compare with similar options?
Compared to the standard Windows Terminal, Intelligent Terminal is an experimental fork adding agent integration; compared to other terminals like Hyper or Alacritty, it offers native agent integration but is Windows-only and heavier.