🧠 Code Next AI

Your Codebase. Your Machine. Your Rules.

A fully offline, privacy-first AI coding assistant that understands your entire codebase — no cloud, no API keys, no subscriptions.

---

🔒 100% Offline. Zero Tokens. Infinite Queries.

Unlike GitHub Copilot, Cursor, or any cloud-based AI tool, Code Next AI runs entirely on your local machine:

Feature	Code Next AI	Cloud AI Tools
🔒 Privacy	Your code never leaves your machine	Code sent to remote servers
💰 Cost	Free forever	API tokens, monthly subscriptions
🌐 Internet	No connection required	Requires internet
⚡ Speed	No rate limits	Rate-limited, throttled
🏢 Enterprise	Works on air-gapped systems	Not possible

---

✨ What It Does

Ask natural language questions about any codebase and get AI-generated answers with referenced source code — all processed locally.

Example questions you can ask:

• "How does the authentication flow work?"
• "Where is the database connection configured?"
• "What does the make_nws_request function do?"
• "Which functions handle error responses?"

---

🏗️ Architecture

Code Next AI uses a state-of-the-art offline RAG (Retrieval-Augmented Generation) pipeline built specifically for source code:

Your Codebase
     │
     ▼
┌─────────────────┐
│  File Scanner   │  Recursively scans files, ignoring .git, node_modules, venv
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│  AST / Tree-    │  Extracts functions & classes with exact line numbers
│  Sitter Parser  │  (Python via ast, JS/TS via tree-sitter)
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│  Chunk Builder  │  Builds semantic CodeChunk objects with metadata
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│ Ollama Embedder │  Generates local vector embeddings (nomic-embed-text)
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│    ChromaDB     │  Persists embeddings in a local vector database
└─────────────────┘
         │  (Query time)
         ▼
┌─────────────────────────────────┐
│        Hybrid Search Engine     │
│  ┌─────────────┐ ┌───────────┐  │
│  │  Semantic   │ │  Keyword  │  │
│  │  Search     │ │  Search   │  │
│  │ (ChromaDB)  │ │ (ripgrep) │  │
│  └──────┬──────┘ └─────┬─────┘  │
│         └──────┬────────┘       │
│                ▼                │
│           Reranker              │
└────────────────┬────────────────┘
                 │
                 ▼
┌─────────────────┐
│  Ollama LLM     │  Synthesizes answer from retrieved context (local model)
└────────┬────────┘
         │
         ▼
   Answer + References (in the UI)

---

🖥️ UI: VS Code–Style 3-Pane Interface

The desktop app (powered by Flet) gives you a familiar IDE-like experience:

• Left Pane — Explorer: Browse your indexed repository files. Click any file to instantly open it in the editor.
• Middle Pane — Code Editor: Full syntax-highlighted code viewer built into the app.
• Right Pane — AI Chat: Ask questions, get streamed answers with collapsible reference panels showing exactly which functions were used.

---

⚙️ Prerequisites

Dependency	Purpose	Install
Python 3.10+	Runtime	python.org
Ollama	Local LLM + Embeddings engine	ollama.com
ripgrep	Keyword search	brew install ripgrep

---

🚀 Quick Start

1. Clone & Install

git clone <repo-url>
cd coding-agent

python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

pip install -r requirements.txt

2. Pull Local Models

Ensure Ollama is running (ollama serve), then pull the required models:

# Embedding model (required for indexing & search)
ollama pull nomic-embed-text

# LLM model for answering (pick one)
ollama pull qwen2.5:7b      # Recommended — fast & capable
ollama pull phi3:mini        # Lighter, faster
ollama pull gemma4:latest    # Google's model

3. Launch the App

python app.py

The desktop UI will open. From there:

1. Click Browse Folder → select your codebase
2. Click Index Repository → wait for indexing to complete
3. Ask your first question in the chat panel!

---

🖥️ CLI Usage (Advanced)

You can also use the agent directly from the terminal:

# Index a repository
python main.py index --repo /path/to/your/project

# Query it
python main.py query --repo /path/to/your/project --query "How does authentication work?"

---

📁 Project Structure

coding-agent/
├── app.py                      # Desktop UI (Flet, 3-pane VS Code layout)
├── main.py                     # CLI entrypoint + core build_index / query_repo
├── settings.json               # Feature flags (log_chunks, log_ast_parser)
├── requirements.txt
│
├── backend/
│   ├── generation/
│   │   └── ollama_llm.py       # Local LLM answer generation
│   └── retrieval/
│       ├── scanner.py          # File discovery & filtering
│       ├── parser.py           # AST (Python) + Tree-Sitter (JS/TS) parsing
│       ├── chunker.py          # CodeChunk model builder
│       ├── embeddings.py       # Ollama nomic-embed-text client
│       ├── vector_store.py     # ChromaDB integration
│       ├── semantic_search.py  # Vector similarity search
│       ├── keyword_search.py   # ripgrep keyword search
│       ├── hybrid_search.py    # Unified search orchestrator
│       └── reranker.py         # Result scoring & ranking
│
└── chroma_db/                  # Local vector database (auto-created)

---

🗺️ Roadmap

• AST-based Python parsing
• Tree-Sitter parsing for JS/TS
• Local embeddings via Ollama
• Hybrid search (semantic + keyword)
• LLM answer generation (fully offline)
• 3-pane VS Code–style desktop UI
• In-app file viewer with syntax highlighting
• Collapsible reference panels in chat
• File watcher for incremental re-indexing on save
• Support for more languages (Go, Rust, Java)
• Multi-repo workspace support
• Chat history persistence

---

🛡️ Privacy Guarantee

Your code is yours. Code Next AI performs all computation locally. No code, query, or result is ever transmitted to any external server.

---

Built with ❤️ for developers who care about their privacy.