WebSharper Code Review

This repository was developed as part of a B.Sc. thesis project in collaboration with IntelliFactory at the University of Dunaújváros (2025). It implements a privacy-first, offline code review extension for F# and WebSharper in Visual Studio Code using local LLMs via Ollama.

Offline AI code review for F# and WebSharper. Runs locally with your Ollama model (e.g., qwen2.5-coder:7b-instruct). Streams diffs, handles large files safely, and snapshots accepted changes into a private Shadow Git history. Turn on RAG to enrich reviews with built-in references packaged in the extension.

Features

• Selection-based reviews with streamed markdown + diff preview.
• Default apply scope (whole file): on typical files, suggestions apply to the entire file.
• Large file safety: for big files (≥ 600 lines), the tool automatically switches to selection-only edits.
• Shadow Git: accepted suggestions are snapshot-committed to a private repo (your real repo is untouched).

• Model switching: change your Ollama model from the Command Palette or on the webview.

• Preferences: adjust the AI's coding style to match your own preferences.

• RAG (toggle): when enabled, reviews are enriched with built-in reference material bundled with the extension. If RAG isn't available on your machine, it's skipped automatically—normal reviews still work.
• RAG status pill (webview): the top bar shows RAG: ON/OFF — click to toggle, then press Refresh to re-run with the new mode.

Quick Start

1. Install the extension.

2. Ensure Ollama service is running locally.

   • Install Ollama:
     Download and run the installer from the official site.

   • Pull the default model (recommended):
     This extension defaults to qwen2.5-coder:7b-instruct. Pull it to get started quickly:

     ollama pull qwen2.5-coder:7b-instruct

   • Prefer a different model?
     Pull your chosen model first, then run the Command Palette action:

     WS Code Review: Change Ollama Model
     

     Select the default endpoint (http://localhost:11434) and enter your model name (exactly as pulled). Do this before running Show Suggestion.

   • Start the service:

     ollama serve

     Default endpoint: http://localhost:11434

   • Verify in your browser: you should see

     Ollama is running
     

3. Activate Show Suggestion

   • Select the code you want reviewed.
   • Run the command:
     • Press Ctrl+Alt+R, or
     • Right-click → WS Code Review: Show Suggestion.

4. (Optional) Use the RAG pill in the top bar to turn RAG ON/OFF, then click Refresh to re-run.

5. Review the streamed suggestion → Accept to apply (and snapshot if Shadow Git is enabled).

Commands

Command	What it does
WS Code Review: Show Suggestion	Requires a selection. For files ≥ 600 lines, only the selected region (plus nearby context and relevant open/module headers) is sent to the AI. For smaller files, the entire file is sent. A streamed diff is shown.
WS Code Review: Change Ollama Model	Pick a different local model (e.g., qwen2.5-coder:7b-instruct).
WS Code Review: Set/Show/Clear AI Preferences	Manage your coding-style preferences used to steer suggestions.
WS Code Review: Show Shadow Git History (Current File)	Browse snapshots made by accepted suggestions.
WS Code Review: Clear Shadow Git History	Purge the private snapshot repo.

Editor context menu: shows on right-click in F# when you have a selection. Keybinding: Ctrl+Alt+R (only in F# with selection).

Settings

All settings live under WS Code Review (Workspace Settings):

• wsCodeReview.git.enable (boolean, default false) Snapshot accepted suggestions into a Shadow Git repo (separate from your real repo).

• wsCodeReview.rag.enable (boolean, default false) Enrich reviews using built-in references packaged with the extension. You can toggle this in Settings or directly via the webview's RAG ON/OFF pill. If RAG is unavailable, it's skipped automatically.

Preferences (coding style)

Use the Set/Show/Clear AI Preferences commands to tailor suggestions to your style—naming, formatting choices, typical WebSharper idioms, and other reviewer hints. Preferences are applied each time you run Show Suggestion.

RAG: Why enable it?

When RAG is on, the assistant receives concise, relevant built-in reference context before reviewing the selection or file. This often yields:

• More project-aware suggestions (naming, idioms, WebSharper patterns).
• Fewer generic rewrites; more precise, domain-specific changes.

Comparison (example):

Without RAG (generic):

With RAG enabled (context-aware):

If RAG cannot run in your environment, reviews continue normally without it.

How it works (high level)

1. Collects your selection + file info (selection is required).
   • For files ≥ 600 lines: builds the prompt from the selected region plus nearby context and relevant open/module lines.
   • For smaller files: builds the prompt from the entire file (even though a selection is required to trigger the command).
2. Streams the AI response into a diff view (line-by-line).
3. On Accept, applies the change and (if wsCodeReview.git.enable=true) creates a Shadow Git snapshot.
   • Apply behavior: files ≥ 600 lines → selection-only apply; smaller files → whole-file apply by default.

Requirements

• VS Code ≥ 1.99
• Ollama running locally with a code model (e.g., qwen2.5-coder:7b-instruct)

Privacy

• No cloud calls by this extension.
• Ollama runs locally.
• Shadow Git history is private and separate; you can clear it anytime.

Troubleshooting

• "Failed to connect to AI"

  • Start the service: ollama serve
  • Verify: open http://localhost:11434 or curl http://localhost:11434/api/tags
  • Ensure the endpoint/model are set via Change Ollama Model

• Model not found

  • Pull the model, e.g.: ollama pull qwen2.5-coder:7b-instruct

• RAG feels the same

  • That's fine; for simple snippets, built-in context may not change much.

• Shadow history not updating

  • Ensure wsCodeReview.git.enable is on and you Accepted a suggestion.

License

See LICENSE.

Third-Party Licenses

This project bundles third-party dependencies whose licenses are listed in THIRD_PARTY_NOTICES.md.