AutoPilot AI

AutoPilot AI is a VS Code extension that adds a sidebar chat assistant for coding tasks. It can answer general questions, explain selected code, refactor snippets, generate documentation, and apply generated code into the active editor.

Features

• Sidebar chat view inside VS Code (AutoPilot AI activity bar item)
• OpenAI-backed chat with API-key sign-in flow
• Context-aware prompting:
  • none: no editor context for general questions
  • selection: selected code context when relevant
  • file: current file context for file-wide questions
• Hybrid request routing:
  • Heuristic classifier
  • Optional mini-model router for ambiguous requests
• Task-aware response depth:
  • quick, normal, deep
• Selection-based editor actions:
  • Explain selection
  • Refactor selection
  • Document selection
• "Apply Code" button on generated code blocks to insert/replace in editor

Requirements

• VS Code ^1.100.0
• Node.js 20+
• An OpenAI Project API key (for example sk-proj-...)

Getting Started

1. Clone and install dependencies:

npm install

2. Compile:

npm run compile

3. Run extension in development:

• Press F5 in VS Code to launch the Extension Development Host.

4. In the extension UI, click Sign In and provide:

• OpenAI Project API key (stored in VS Code global state)

Commands

The extension contributes these commands:

• autopilot-searchbar.search — Open AutoPilot AI
• autopilot-searchbar.logout — Logout
• autopilot-searchbar.explainSelection — Explain with AutoPilot AI
• autopilot-searchbar.refactorCode — Refactor Code with AutoPilot AI
• autopilot-searchbar.documentSelection — Document It with AutoPilot AI
• autopilot-searchbar.applyCode — Apply Code from AutoPilot AI

Selection commands are also available in the editor context menu when text is selected.

Extension Settings

This extension contributes the following settings:

• autopilot.apiKey (string): API key field (primary sign-in uses secure global state flow).
• autopilot.enableAiRouter (boolean, default true): Enables mini-model request classification when heuristics are uncertain.
• autopilot.routerModel (string, default gpt-4o-mini): Model used for the lightweight request router.

How Context Routing Works

For normal chat messages, the extension plans each request before sending it to the main model:

1. Detect intent/context need (none / selection / file)
2. Detect response depth (quick / normal / deep)
3. Build minimal prompt payload
4. Send to OpenAI

This reduces unnecessary token usage and improves latency versus always attaching full-file context.

Development Scripts

• npm run compile — Build extension
• npm run watch — Build in watch mode
• npm run package — Production bundle
• npm run lint — Lint source
• npm run check-types — Type-check only
• npm run test — Run extension tests

Packaging and Publishing

1. Prepare package.json

Ensure at least the following are set:

• name
• displayName
• publisher
• version
• engines.vscode

2. Create a VSIX package

npx vsce package

This generates a file like:

• autopilot-ai-<version>.vsix

3. Publish to Visual Studio Marketplace

1. Create/manage publisher: https://marketplace.visualstudio.com/manage
2. Create Azure DevOps PAT with Marketplace publish permissions
3. Publish:

npx vsce publish -p <YOUR_PAT>

Troubleshooting

• If authentication fails, verify you are using a valid OpenAI Project API key.
• If responses are empty due to token limits, ask a narrower question or reduce file/selection size.
• If network errors occur, check proxy/firewall settings in your environment.

Security Notes

• API key is read from VS Code global state in this extension.
• Review your organization policy before using personal or shared keys.

License

See LICENSE (add one before marketplace publish if missing).