CodeTorch

AI-generated natural language outlines help you understand code faster, with less effort, and skim code for easier navigation.

Features

Annotations: Outlines are rendered as annotations to stay unobtrusive even when outlines are extensive and so your source code isn’t touched.

Live Documentation: Outlines update in the background as you save your work.

Reviewing AI-generated Code: Outlines overlay AI-generated diffs for easier review.

Storing Documentation: Outlines are saved in your project’s root (/.codetorch) and can be pushed to GitHub for re-use by collaborators.

Requirements

Setting up a Gemini API Key

CodeTorch uses Gemini-2.0-Flash to generate code summaries. You must configure your Google AI API Key to use the extension:

1. Open "Preferences: Open User Settings (JSON)": Cmd/Ctrl + Shift + P
2. Add the following field to the JSON: "codetorch.googleApiKey": "your-actual-gemini-api-key-here"

Need to create an Google AI API key? Create one in a few clicks and use free credits for the first 90 days (July 2025).

Setting up outline sytling

Extension creators don't have access to CodeLens annotation styling, so users must manually configure it. Configure them in "Preferences: Open User Settings (JSON)" (Cmd/Ctrl + Shift + P).

Here are recommend styling fields to add:

"editor.codeLens": true,
"workbench.colorCustomizations": {
    // For dark color themes
    "editorCodeLens.foreground": "#a7a7a7", 
    // For light color themes
    // "editorCodeLens.foreground": "#666666"
},
"editor.codeLensFontSize": 12,
"editor.codeLensFontFamily": "Helvetica"

Extension Settings

Triggering outline re-generatoin

By default, oultines are re-generated everytime you save a file. You can alternatively configure a keyboard shortcut to trigger re-generation.

1. In User Settings (JSON), add "codetorch.regenerationTrigger": "manual"
2. In Keyboard Shortcuts, search and configure "Regenerate Summaries"

References

Thanks to the prior work this project is built on:

1. Shi, Kensen, Deniz Altınbüken, Saswat Anand, Mihai Christodorescu, Katja Grünwedel, Alexa Koenings, Sai Naidu et al. "Natural language outlines for code: Literate programming in the llm era." In Companion Proceedings of the 33rd ACM Symposium on the Foundations of Software Engineering (FSE '25), June 23--27, 2025, Trondheim, Norway
2. Su, C.Y., Bansal, A., Huang, Y., Li, T.J.J. and McMillan, C., 2024. Context-aware Code Summary Generation. arXiv preprint arXiv:2408.09006.