Zed Editor

Zed Agent is Zed's native agentic workflow. It operates within the Agent Panel and Threads Sidebar, integrating directly with the editor, project files, terminal, and review surfaces.

Use Zed Agent when you need an agent to perform the following tasks:

• Project Awareness: Read and search through your project files.
• File Manipulation: Edit files directly.
• Command Execution: Run terminal commands.
• Tool Integration: Use Zed-managed MCP (Model Context Protocol) tools.
• Contextual Guidance: Follow specific Agent Profiles, Skills, and Instructions.
• Visual Feedback: Show code changes within Zed's native review UI.

PowerShell language support in Zed is provided by the community-maintained Zed PowerShell extension.

Key components used:

• Tree-sitter: airbus-cert/tree-sitter-powershell for syntax highlighting.
• Language Server: PowerShell/PowerShellEditorServices for LSP features like autocompletion and diagnostics.

Zed's AI capabilities are categorized into three primary areas:

1. Agents: Defines how agentic work is executed (e.g., Zed Agent, External Agents, or Terminal Threads).
2. Model Access: Defines how Zed connects to language models (e.g., Zed-hosted models, API access, subscriptions, gateways, or local models).
3. Features: The specific AI workflows available to the user (e.g., Agentic editing, inline edits, edit prediction, or Git assistance).

Use the AI Quick Start if you have a specific task in mind, or AI by Company if you want to configure a specific provider or subscription.

GPUI provides several specialized services to handle common application requirements:

• Actions: User-defined structs used to map keystrokes to logical UI operations (e.g., implementing cmd-q). See the action module.
• Platform Services: Methods on app::App to perform system-level tasks like quit the app or open a URL.
• Async Executor: An executor integrated with the platform's event loop. See the executor module.
• Testing: The [gpui::test] macro allows for testing GPUI applications. Tests use a TestAppContext to simulate platform input. See app::test_context and test modules.

Zed's built-in AI agent uses a set of tools to interact with your codebase, the web, and your terminal. These tools allow the agent to read, search, edit, and execute commands.

Key concepts for managing tools:

• Availability: The specific tools available depend on your Agent Profile, the selected model provider, and your Zed version.
• Permissions: You can configure whether tool actions are automatically approved, automatically denied, or require manual confirmation via Tool Permissions.
• Customization: To extend the agent's capabilities beyond built-in tools, you can connect MCP servers.
• Sandboxing: The terminal tool can be subject to OS-level restrictions when Zed Agent sandboxing is enabled.

Zed's code editing experience is built on the integration of language servers, AI-powered predictions, and multi-file editing capabilities. The core workflow combines:

1. Language Servers: Provide the foundation for completions, diagnostics (errors/warnings), and quick fixes based on project types and structure.
2. Edit Predictions: Suggest multi-character or multi-line changes in real-time as you type.
3. Multibuffers: Enable simultaneous editing across multiple files using multiple cursors.

This synergy allows for complex workflows, such as performing a rename refactor via a language server and immediately seeing/editing the impact across all affected files in a multibuffer.

Luau support in Zed is provided by the community-maintained Luau extension.

Key components used for language support:

• Tree-sitter: 4teapo/tree-sitter-luau for syntax highlighting.
• Language Server: JohnnyMorganz/luau-lsp for LSP features.

For detailed configuration instructions, refer to the Luau Zed Extension README.

Zed includes four primary systems for executing and inspecting code:

1. Terminal: A built-in emulator that supports multiple sessions and shell customization. It is the environment where tasks and commands execute.
2. Tasks: A system for defining and running shell commands. Tasks can access editor context, such as the current file, text selection, or symbols, making them ideal for builds, linting, or repeatable scripts.
3. Debugger: A built-in debugger that uses the Debug Adapter Protocol (DAP). It supports setting breakpoints, stepping through code, and variable inspection for languages including C, C++, Go, JavaScript, Python, Rust, and TypeScript.
4. REPL: An interactive execution environment using Jupyter kernels. It allows for inline execution of code selections or cells, supporting languages like Python, TypeScript (via Deno), R, and Julia.

Zed features an extension system that allows users to add new functionality to the editor. Extensions can be used to provide:

• Themes: Custom visual styles for the editor UI.
• Language Support: Syntax highlighting, language server protocol (LSP) integration, and other language-specific features.
• AI Tools: Integration with various AI models and workflows.
• Icon Themes: Custom icons for files and directories.
• Debugger Extensions: Support for debugging different programming languages.
• Snippets: Code completion templates.
• MCP Servers: Model Context Protocol (MCP) server integration to extend AI capabilities.

Zed provides several distinct AI-powered workflows:

• Agent Panel: Used to prompt agents, add context, review changes, and manage active threads.
• Parallel Agents: Enables running multiple threads across different projects and worktrees.
• Inline Assistant: Allows for transforming a text selection in place.
• Edit Prediction: Provides AI-driven completions that you can accept while typing.
• Git commit generation: Automatically generates commit messages from the Git panel.