Agentic GitHub Code Reviewer

A fully local, privacy-safe multi-agent system that automatically reviews
any GitHub repository and generates a detailed code review report powered by LangGraph + LM Studio.
No API keys. No data leaves your machine.

---

What It Does

Give it any GitHub repository URL → it automatically:

• Fetches all source files via GitHub API
• Indexes code into a ChromaDB vector store (RAG)
• Reviews each file for bugs, security issues, performance problems
• Suggests minimal, targeted code fixes
• Generates a full markdown report with executive summary + score

---

Architecture

                        GitHub URL
                            ↓
┌──────────────────────────────────────────────────────┐
│              LangGraph State Machine                 │
│                                                      │
│  [FETCHER] → [REVIEWER] → [SUGGESTER] → [SUMMARISER] │
│      ↓            ↓            ↓             ↓       │
│  GitHub API    RAG Query    Fix Suggest   .md Report │
└──────────────────────────────────────────────────────┘
                            ↓
                 review_YYYYMMDD_HHMMSS.md

---

Agent Roles

Agent	Responsibility
Fetcher	Calls GitHub API, extracts code from notebooks, builds ChromaDB RAG index
Reviewer	Static analysis:- bugs, security, performance, readability
Suggester	Generates before/after code fixes
Summariser	Writes executive summary with verdict + score

---

Tech Stack

Layer	Technology
Agent Orchestration	LangGraph
LLM Framework	LangChain
Local LLM	LM Studio
Embeddings	Nomic Embed Text
Vector Store	ChromaDB
Web UI	Streamlit
REST API	FastAPI

---

Prerequisites

• Python 3.11+
• LM Studio installed and running
• GitHub Personal Access Token
• Chat model loaded in LM Studio
• Embedding model loaded

---

Setup & Installation

1️. Clone Repo

git clone https://github.com/Ravevx/Agentic-GitHub-Code-Reviewer.git
cd Agentic-GitHub-Code-Reviewer

2️. Create Environment

conda create -n agent-local python=3.11
conda activate agent-local

3️. Install Dependencies

pip install -r requirements.txt

4️. Create .env

GITHUB_TOKEN=your_token_here

5️. Configure Model

Edit config.py

LM_MODEL = "your-model-name"
MAX_FILE_SIZE = 4000
MAX_FILES = 5

---

Usage

CLI

python main.py

---

Web UI

streamlit run app.py

---

API

uvicorn api:app --reload

Docs → http://localhost:8000/docs

---

Example Output

   AI CODE REVIEW AGENT

[FETCHER] Fetching repository files from GitHub...
Repo: Machine-Learning
Loading 9 code files...
[FETCHER] Loaded 9 files

[RAG] Building code index...
Indexed 63 chunks from 9 files
[RAG] Code index ready!

[REVIEWER] Analysing code...
Reviewing: decision_tree.ipynb
Reviewing: logistic_regression.ipynb
[REVIEWER] Reviewed 9 files

[SUGGESTER] Generating fix suggestions...
[SUGGESTER] Generated suggestions for 9 files

[SUMMARISER] Writing final report...
[SUMMARISER] Report complete!

Report saved to: review_20260220_153700.md

Report output:

VERDICT: Approved with Minor Fixes
SCORE: 8.5/10
SECURITY RISK: Low

TOP ISSUES
• `train_claim_type.py` Data Leakage Risk
   - Line: `self.df = self.df[self.df["claim"] == 1].reset_index(drop=True)`
• `model_span.py` Dimension Mismatch
  - Line: `self.classifier = nn.Linear(hidden_size, 1)`
• `train_span.py` Unsafe Hardcoded Splits

---

Project Structure

i-Agentic-GitHub-Code-Reviewer/
│
├── main.py                  ← CLI runner
├── api.py                   ← FastAPI REST server
├── app.py                   ← Streamlit web UI
├── llm.py                   ← LM Studio LLM factory
├── config.py                ← Settings (URL, model, limits)
├── test_lm.py               ← LM Studio connection test
├── requirements.txt         ← Dependencies
├── .env                     ← GitHub token (not committed)
│
├── tools/
│   └── github_tools.py      ← GitHub API integration
│
├── rag/
│   └── code_store.py        ← ChromaDB vector index
│
├── graph/
│   └── review_graph.py      ← LangGraph pipeline
│
└── agents/
    ├── fetcher.py            ← Agent 1: Fetch + RAG
    ├── reviewer.py           ← Agent 2: Code review
    ├── suggester.py          ← Agent 3: Fix suggestions
    └── summariser.py         ← Agent 4: Final report

---

Config Reference

Setting	Default
MAX_FILE_SIZE	4000
MAX_FILES	5
LM_MODEL	ministral-3b
Context	8192

---

Troubleshooting

Problem	Fix
Model won't connect	Start LM Studio
Timeout	Reduce file limits
Token error	Increase Tokens

---

License

MIT License

---

Credits

• LangGraph
• LM Studio
• ChromaDB
• GitHub API