Code review that knows your rules.
What is it?
Revvy is a VS Code extension that brings structured, AI-powered code review directly into your editor. Instead of generic AI feedback, it enforces a set of rules you define in YAML — tailored to your team's standards, domain, or industry requirements (e.g. safety-critical embedded C/C++, Python, and more).
Why not just use /review or type a prompt?
- Many developers are not good at writing prompts.
- Every review requires crafting a new prompt tailored to a specific tech stack or language.
- AI can't deliver a thorough review without specific rules or best practices — the kind that are usually defined at the company or team level.
- Rules vary from one developer to another, which is why Revvy uses hardcoded, shared rules stored in a single YAML file that the whole team uses.
- A clean UI that shows you the exact line of code with an inline comment when an issue or warning is detected.
That's all?
Not quite. Revvy also:
- Reviews multiple GitHub PRs and GitLab MRs in a single pass using MCP
- Validates your code against Jira ticket requirements — so nothing slips out of scope
- Generates integration tests (functional + regression)
- Provides a Copy as Markdown button that copies all review comments along with a ready-to-use prompt — paste it into your favorite AI assistant (ChatGPT, Claude, Copilot, etc.) to fix every issue in one shot
- (Coming soon) Auto-comment on PRs/MRs — Revvy posts a comment on the pull/merge request stating it has been reviewed by the tool, with a summary of findings for reviewers and maintainers to see at a glance
It works with the AI backend you already have — GitHub Copilot (no extra key needed), OpenAI GPT-4o, or Anthropic Claude — with automatic fallback.
Review Modes
Revvy offers two review modes you can toggle in the panel. Each is optimized for a different MR shape.
Per File mode (default)
Each file is reviewed in its own AI request, in parallel. Use for complex MRs with many files or substantial changes.
flowchart LR
MR[Merge Request5 files modified]
MR --> S1[Split by file]
S1 --> F1[file1.cpp]
S1 --> F2[file2.h]
S1 --> F3[file3.cpp]
S1 --> F4[file4.h]
S1 --> F5[file5.cpp]
F1 --> AI1["AI request 1(1 premium request)"]
F2 --> AI2["AI request 2(1 premium request)"]
F3 --> AI3["AI request 3(1 premium request)"]
F4 --> AI4["AI request 4(1 premium request)"]
F5 --> AI5["AI request 5(1 premium request)"]
AI1 --> R["Merged reviewTotal: 5 requests"]
AI2 --> R
AI3 --> R
AI4 --> R
AI5 --> R
What it solves: focused attention per file, no context dilution, no output truncation. Each finding is reasoned about with the full attention budget of the model.
Cost: higher — N requests for N files. Best quality on complex MRs.
All in One mode
The entire MR is sent to the AI in a single request. Use for small MRs (under ~5 files / ~300 lines of diff).
flowchart LR
MR[Merge Request3 small files]
MR --> C[Combined diff]
C --> AI["Single AI request(1 premium request)"]
AI --> R["ReviewTotal: 1 request"]
What it solves: lower quota consumption when the MR is small enough that focused per-file attention isn't necessary. Useful for small fixes, opcode additions, or single-feature changes.
Cost: 1 request total. May miss findings on larger MRs.
Use All in One for small MRs (under ~5 files / ~300 lines of diff).
Use Per File for larger or complex MRs.
Demo
Quick Start
From the VS Code Marketplace:
Search for Revvy in the VS Code Extensions panel and install it.
From source:
git clone https://github.com/YonK0/revvy.git
cd revvy
npm install
npm run compile
# Press F5 in VS Code to launch the Extension Development Host
Once installed, click the Revvy icon in the Activity Bar. Three example profiles are created automatically on first use (c-embedded, yocto, python).
Writing Rule Profiles
Create .yaml files in .vscode-reviewer/profiles/. Rules are reloaded automatically on save.
profile:
id: my-team
label: "My Team Rules"
version: "1.0.0"
file_patterns:
- "**/*.ts"
rules:
- id: NO_CONSOLE_LOG
title: "No console.log in production code"
category: "coding-standards"
severity: warning # error | warning | info
enabled: true
description: |
console.log statements should not be merged to main.
Use a structured logger instead.
See the bundled profiles in .vscode-reviewer/profiles/ for full examples.
MCP Servers (Optional)
Enables Jira ticket fetching and multi-repo PR/MR review. Copy .vscode/mcp.json and fill in your credentials — tokens are prompted at runtime and never stored on disk.
| Server |
Install |
Used for |
| GitHub |
Docker |
Fetch PR diffs for multi-repo review |
| GitLab |
npm i -g @modelcontextprotocol/server-gitlab |
Fetch MR diffs |
| Atlassian |
pip install mcp-atlassian |
Fetch Jira ticket requirements |
Requirements
- VS Code
1.96.0+
- Git in
PATH
- One of: GitHub Copilot subscription or free tier / OpenAI API key / Anthropic API key
Architecture
flowchart LR
User["👤 User"] --> Trigger["Trigger review(local diff or remote MR)"]
Trigger --> Engine["Revvy engine"]
Rules["📄 YAML rules"] -.-> Engine
Engine --> AI["AI backendCopilot · OpenAI · Anthropic"]
AI --> Out["Findings in VS Code"]
Code Architecture
src/
├── extension.ts # Entry point — commands, WebView provider, YAML watcher
├── panelProvider.ts # WebView UI — welcome, loading, results screens
├── reviewer.ts # Core engine — prompt construction, AI call, response parsing
├── aiBackend.ts # AI abstraction — Copilot, OpenAI, Anthropic + fallback chain
└── ruleLoader.ts # YAML loader — file scanning, parsing, live-reload watcher
Contributing
Contributions are welcome — new rule profiles, AI backends, bug fixes, UI improvements. See CONTRIBUTING.md.
