Code Analyzer

A lightweight Flask web app that analyzes code files or entire project folders using a local AI model. Upload your code, get back a structured review with severity-labeled issues, architecture summary, and syntax-highlighted suggestions — all running locally with no external API calls.

---

Features

• Folder or file upload — analyze a single file or an entire project directory
• Two-pass analysis — pass 1 summarizes architecture, pass 2 does a deep review using that context
• Severity labels — issues are labeled [CRITICAL] [HIGH] [MEDIUM] [LOW] and color-coded
• Line number references — line numbers are injected into context so the AI can pinpoint exactly where issues are
• Proper file tree — uploaded project structure is rendered as a real ASCII tree
• Syntax highlighting — code blocks in AI output are highlighted via highlight.js
• Summary caching — re-analyzing the same project skips pass 1 (cached by content hash)
• Works with any OpenAI-compatible API — Ollama, LM Studio, or any local model server

---

Requirements

• Python 3.10+
• A running local model server (Ollama, LM Studio, etc.)

---

Setup

# Clone the repo
cd code-analyzer

# Install dependencies
pip install flask requests

# Start your model server (example with Ollama)
ollama serve
ollama pull llama3.2:3b

# Run the app
python app.py

Open http://localhost:5000 in your browser.

---

Configuration

Edit the config block at the top of app.py:

API_URL            = "http://localhost:11434/v1/chat/completions"
MODEL              = "dolphin3.0-llama3.1-8b"

MAX_FILES          = 50
MAX_CHARS_PER_FILE = 8000
MAX_TOTAL_CHARS    = 25000

Setting	Description
API_URL	Any OpenAI-compatible /v1/chat/completions endpoint
MODEL	Model name as your server expects it
MAX_FILES	Max files accepted per upload
MAX_CHARS_PER_FILE	Per-file character limit before truncation
MAX_TOTAL_CHARS	Total context limit (~6k tokens, safe for 8k context models)

---

Supported File Types

.py .js .ts .jsx .tsx .html .css .scss .java .c .cpp .h .go .rs .rb .php .json .yaml .yml .toml .sh .bash .md .txt

---

Project Structure

code-analyzer/
├── app.py
└── templates/
    └── index.html

---

How It Works

1. Files are uploaded and line numbers are prepended to each file (0001: ...)
2. A proper ASCII file tree is built from the uploaded paths
3. Pass 1 — the AI summarizes the project architecture (result is cached)
4. Pass 2 — the AI does a deep review, grounded by the summary and file tree
5. Results are rendered with severity coloring and syntax highlighting

---

Planned (future)

• Chat tab for follow-up questions about the review
• Docker container for integration with existing stacks
• Git repo URL input (clone and analyze directly)
• Cross-project analysis via vector search when integrated with Qdrant

---

Notes

• The summary cache is in-memory and resets when the server restarts
• Drag and drop works for files; folder drag-and-drop behavior varies by browser — use the folder picker button for reliability
• Works best with code-focused models (e.g. deepseek-coder, codellama, qwen2.5-coder) but any capable model will work
• Firefox will not work with drag and drop, use file picker.