VSCode Aider

Enables developers to right-click on code selections within the editor and invoke AI-assisted refactoring through Aider's backend, which parses the selected code, sends it to OpenAI/Anthropic APIs, and streams back refactored code that can be applied directly to the file. The extension maintains bidirectional sync between VS Code's editor state and Aider's session state, ensuring file modifications persist across both interfaces.

When developers right-click on code errors (syntax, runtime, or linting errors) in VS Code, the extension extracts error metadata (error message, line number, error type) and sends it along with surrounding code context to the configured AI model. The AI generates fix suggestions that are streamed back and can be applied inline, with the extension maintaining awareness of which errors have been addressed.

The extension stores configuration in VS Code's settings system (settings.json), persisting user preferences for default model, API keys, and custom Aider CLI arguments across sessions. Settings are scoped to the workspace or user level, allowing team-wide configuration via .vscode/settings.json or individual customization. The extension reads settings on startup and applies them to all subsequent operations.

Developers can invoke the `Aider: Select Model` command from the VS Code command palette to switch between supported AI models (GPT-4, Claude, and undocumented 'new additions') without restarting the extension or Aider CLI. The selection is persisted in extension settings and applied to all subsequent AI operations in the current session, with the status bar displaying the currently active model.

The extension provides a `Aider: Generate README.md` command that sends the project's file structure, key files, and metadata to the configured AI model, which generates a comprehensive README.md file with sections for installation, usage, and architecture. The generated file is written to the project root and can be edited or regenerated, with the extension tracking whether a README already exists to avoid overwriting.

The extension provides file explorer context menus to add or ignore files from the Aider session, maintaining a persistent list of tracked files. It synchronizes this state bidirectionally with the Aider CLI tool, ensuring that files modified in VS Code are reflected in Aider's session and vice versa. The extension tracks open files on startup but may miss some files, requiring manual re-sync via the file explorer.

The extension adds a clickable status bar item at the bottom of VS Code that displays the currently active AI model and provides quick access to Aider operations. Clicking the status bar item opens a menu or launches Aider, and the item updates in real-time to reflect the selected model, providing visual feedback without requiring command palette access.

The extension registers multiple commands in VS Code's command palette (accessible via Ctrl+Shift+P) including `Aider: Open`, `Aider: Select Model`, and `Aider: Generate README.md`. These commands provide keyboard-driven access to core Aider operations without requiring mouse interaction or menu navigation, with command names discoverable via fuzzy search in the palette.

The extension abstracts API interactions with both OpenAI (GPT-4) and Anthropic (Claude) through a unified interface, handling authentication, request formatting, and response streaming. It stores API keys in VS Code's secure credential storage and routes requests to the appropriate API based on the selected model, with support for undocumented 'new additions' suggesting a pluggable model architecture.

The extension manages the lifecycle of the external Aider CLI process, including launching, monitoring, and terminating it based on user actions. It handles process output streaming, error detection, and restart logic, with the `Aider: Open` command spawning a new CLI session or connecting to an existing one. The extension maintains awareness of CLI state and prevents duplicate processes.

The extension enables developers to request new code generation (functions, classes, modules) with full awareness of the project's file structure and existing code patterns. It sends the current file, selected code context, and optionally other tracked files to the AI model, which generates code that follows the project's conventions and integrates seamlessly. The generated code is streamed back and can be inserted at the cursor position or reviewed before applying.