Revisor is a local, privacy-first AI assistant that helps developers understand incoming commits before merging them. Controlled directly from your terminal with a single command, Revisor fetches the latest remote changes, analyzes the diff against your local branch, and runs it through a local AI model to generate a structured review. The final report is displayed instantly inside VS Code as a side-by-side Markdown document.
- Terminal Workflow: Trigger code reviews straight from your regular terminal using a native Git alias.
- Local & Private: Your source code never leaves your machine. All analysis is performed entirely on your local hardware.
- Clean Pre-Merge Sight: Automatically executes a
git fetchbackground pipeline to see what has changed on the remote branch without polluting your current working state. - Rich Markdown Reports: Displays a clean review panel highlighting summaries, breaking changes, and structural risks.
Once installed, Revisor injects a lightweight global script wrapper into your Git configuration.
Before your first review, you will need to select your preferred AI model:
- Open the VS Code Command Palette (
Ctrl+Shift+P). - Search for "Revisor: Switch AI Model".
Revisor will automatically provision a dedicated custom model based on your selection to handle the report generation. You can change this active model at any time.
When you run git revisor in your terminal, the extension uses a local file-bridge pipeline (pending-args.txt) to securely pass environment metrics from your shell directly to the VS Code Extension Host, seamlessly bypassing traditional CLI-to-editor limitations.
Revisor relies on Git and a local AI engine to guarantee complete data security.
- Install Ollama: Download and run Ollama on your machine.
- Pull a Coding Model: Open your terminal and download a model optimized for code intelligence. Depending on your machine's hardware, we recommend:
- Lightweight (8GB RAM, no dedicated GPU):
ollama pull qwen2.5-coder:1.5b(Note: The 1.5B parameter limit restricts deep reasoning. Use this primarily for simple, everyday changes rather than complex architectural reviews.) - Standard (16GB RAM, 6GB+ VRAM):
ollama pull deepseek-coder-v2:lite - High performance (32GB RAM, 10GB+ VRAM):
ollama pull qwen2.5-coder:14b
- Lightweight (8GB RAM, no dedicated GPU):
- Start Ollama: Run
ollama servein your terminal.
Open your terminal in any active Git repository and type:
git revisor <remote> <localBranch> <remoteBranch>
- Review quality depends on the model selected. Smaller models (1.5b) may produce shallow or inconsistent analysis. For best results use
qwen2.5-coder:7bor larger. - The extension requires Ollama to be running before VSCode starts. If you start VSCode before
ollama serve, restart the extension host viaDeveloper: Restart Extension Hostfrom the command palette. - Currently built and optimized for Windows. MacOS and Linux support are not tested.
- Very large diffs (500+ lines) are trimmed before being sent to the model to avoid context window limits.
Initial release of Revisor.
Updated logo and minor inconsistency fixes
Thank you for choosing Revisor!