π AI-Native Local review tool with GitHub-Style Interface
HighReviewλ λ‘컬 νκ²½μ λ°©ν΄νμ§ μμΌλ©΄μ GitHub Pull Requestλ₯Ό κ°λ ₯νκ² λ¦¬λ·°ν μ μλ λꡬμ λλ€. Git Worktreeλ₯Ό νμ©νμ¬ νμ¬ μμ λλ ν 리λ₯Ό 건λλ¦¬μ§ μκ³ , Offline-First μ½λ λΆμκ³Ό Local AIλ₯Ό ν΅ν΄ μμ νκ³ κ°λ ₯ν 리뷰 νκ²½μ μ 곡ν©λλ€.
π λ‘κ·ΈμΈ λΆνμ: HighReviewλ μ¬μ©μμ λ‘컬
ghν΄λΌμ΄μΈνΈμ AI μμ΄μ νΈλ₯Ό μ§μ νμ©νλ―λ‘, λ³λμ νμκ°μ μ΄λ μΈμ¦ μ μ°¨ μμ΄ μ¦μ μ¬μ©ν μ μμ΅λλ€.
HighReview is a powerful local PR review tool that doesn’t disrupt your working environment. Using Git Worktree, it provides IDE-level code analysis and AI insights without touching your current working directory.
π No Login or Authentication Required: HighReview operates entirely locally using your existing
ghclient and local AI agents. No separate account or sign-up is needed.
View on GitHub | Report Bug | Request Feature
- Git Worktree Integration: Creates isolated review environments instantly.
- Background Indexing: Pre-indexes symbols for instant code navigation.
- IDE-Like Experience: Go to Definition, Find Usages, and Symbol Search (TypeScript, Ruby, Java).
Plug-and-play AI support. No API keys required for local models.
- Supported Providers:
- Claude Code CLI (Recommended, Anthropic)
- Ollama (Local Llama 3, Mistral, etc.)
- LM Studio (Local, OpenAI Compatible)
- Automatic Detection: Automatically detects installed providers.
- Zero Configuration: Select your provider in the UI Settings.
Ask questions about your code with full context awareness.
- Smart Context: Automatically includes selected code, file contents, and PR details.
- Deep Analysis: “Explain this function”, “Find bugs in this selection”, “Generate tests”.
- Multi-Modal: Supports file attachments and reference links.
- Automatic Issue Detection: Critical bugs, warnings, and optimization suggestions.
- Change Intent Analysis: Understand why code was changed.
- Impact Analysis: See potential side effects and breaking changes.
- Visual Call Stacks: Interactive Mermaid flowcharts and sequence diagrams.
HighReview/
βββ apps/
β βββ cli/ # Node.js Backend (Fastify + SQLite + Tree-sitter)
β βββ web/ # React Frontend (Vite + Monaco + React Query)
βββ .highreview/ # Global Config & Storage
flowchart TD
%% Nodes
Start(["User Initiates Review"])
subgraph Stage1 ["Stage 1: Context Optimization"]
Diff["Get Git Diff"]
TreeSitter["ContextAnalyzer
(Tree-sitter)"]
OptContext["Optimized Context
(Signatures & Docstrings)"]
Diff --> TreeSitter
TreeSitter --> |Extract| OptContext
end
subgraph Stage2 ["Stage 2: Map Phase (Batch Processing)"]
Chunking["ChunkingStrategy
(Split Files)"]
Batch1["Batch 1 Review
(Code Reviewer Persona)"]
Batch2["Batch 2 Review
(Code Reviewer Persona)"]
BatchN["Batch N Review
(Code Reviewer Persona)"]
OptContext --> Chunking
Chunking --> Batch1
Chunking --> Batch2
Chunking --> BatchN
end
subgraph Stage3 ["Stage 3: Reduce Phase (Global Analysis)"]
Collect["Collect & Merge Results"]
Architect["PR Architect AI
(RefineSummaryWithAI)"]
Output1["Global Change Intent"]
Output2["Impact Analysis"]
Output3["Logic Flow Diagrams
(Mermaid)"]
end
%% Edge Connections
Start --> Diff
Batch1 --> |Issues + 1-Line Summary| Collect
Batch2 --> |Issues + 1-Line Summary| Collect
BatchN --> |Issues + 1-Line Summary| Collect
Collect --> |All Summaries| Architect
Architect --> Output1
Architect --> Output2
Architect --> Output3
Output1 --> Final(["Final Review Report"])
Output2 --> Final
Output3 --> Final
%% Styling
classDef stage fill:#e1f5fe,stroke:#01579b,stroke-width:2px;
classDef process fill:#fff9c4,stroke:#fbc02d,stroke-width:2px;
classDef ai fill:#e8f5e9,stroke:#2e7d32,stroke-width:2px,rx:10,ry:10;
class Stage1,Stage2,Stage3 stage;
class TreeSitter,Chunking,Collect process;
class Batch1,Batch2,BatchN,Architect ai;
Loading
Backend (CLI)
- Runtime: Node.js 18+
- Framework: Fastify
- Database: SQLite (
better-sqlite3) for robust local storage. - Analysis: Tree-sitter (Java, TS, Ruby, Python, Kotlin) for precise code parsing.
- AI Integration: Custom Provider Architecture (Claude, Ollama, LM Studio).
Frontend (Web)
- Framework: React 18, Vite
- Editor: Monaco Editor (VS Code core)
- State: React Query, Zustand
- UI: Allotment (Splits), TailwindCSS, Radix UI
- Visualization: Mermaid.js for flows and charts.
- Node.js: >= 18.0.0
- Git: >= 2.30.0
- GitHub CLI:
gh auth loginrequired for PR fetching. - AI Provider (Optional but Recommended):
-
Clone the repository
git clone https://github.com/HighGarden-Studio/HighReview.git cd HighReview -
Install dependencies
-
Start Development Server
# Starts both CLI (port 8765) and Web (port 5273) npm run dev -
Open in Browser
- Go to
http://localhost:5273 - Settings Setup: Navigate to the Settings page to select your AI Provider.
- Go to
Review-requested PRs are automatically loaded. Click any card to start a workspace-isolated review.
When re-running a review, you can customize the analysis depth:
- π§ Context-Aware Review: Uses Tree-sitter to find callers and implementations of modified code in the project.
- π― Change Intent Analysis: Determines why code was changed (File-level or Block-level).
- π Code Visualization: Generates Mermaid Flowcharts or Sequence Diagrams or Class Diagrams for modified functions.
- π Impact Analysis: Assesses broader impact (Module, Project, or Dependencies).
- β¨ Semantic Diff: Detects moved code, refactoring patterns, and ignores whitespace/comments.
- π¬ Custom Prompt: Add your own specific instructions for the AI reviewer.
- Auto-Review: The AI analyzes the PR on load. Click “Re-run” to trigger a fresh analysis with custom options.
- Ask Assistant: Highlight any code in the specialized “Diff Editor” or “File Tree” and ask questions in the side panel.
- Go to Definition: Right-click on symbols (even in Ruby/Java) to jump to their definition.
- Find Usages: See where functions or classes are used across the PR.
This project is licensed under the Apache License 2.0 – see the LICENSE file for details.
Made with β€οΈ by HighGarden Studio








