A fully offline, privacy-first AI coding assistant that understands your entire codebase — no cloud, no API keys, no subscriptions.
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 |
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_requestfunction do?" - "Which functions handle error responses?"
Code Next AI uses a state-of-the-art offline RAG (Retrieval-Augmented Generation) pipeline built specifically for source code:
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.
| Dependency | Purpose | Install |
|---|---|---|
| Python 3.10+ | Runtime | python.org |
| Ollama | Local LLM + Embeddings engine | ollama.com |
| ripgrep | Keyword search | brew install ripgrep |
git clone <repo-url>
cd coding-agent
python -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
pip install -r requirements.txtEnsure 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 modelpython app.pyThe desktop UI will open. From there:
- Click Browse Folder → select your codebase
- Click Index Repository → wait for indexing to complete
- Ask your first question in the chat panel!
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?"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 # Weaviate integration
│ ├── semantic_search.py # Vector similarity search
│ ├── keyword_search.py # ripgrep keyword search
│ ├── hybrid_search.py # Unified search orchestrator
│ └── reranker.py # Result scoring & ranking
│
- 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
Your code is yours. Code Next AI performs all computation locally. No code, query, or result is ever transmitted to any external server.