Inline - Offline Code Completion

Inline is a privacy-first VS Code extension that delivers intelligent, AI-powered code completion entirely offline. All processing happens locally on your machine—your code never leaves your device.

---

✨ Features

Core Features

Model Management

• One-click downloads for recommended models (DeepSeek, CodeLlama, Phi-3)
• Drag & drop .gguf file import
• URL-based model downloads from Hugging Face
• Automatic model validation and metadata detection

Code Intelligence

• Completions: Context-aware inline suggestions
• Hover Info: Type information and documentation
• Code Actions: AI-powered quick fixes and refactoring
• Diagnostics: Error detection and vulnerability scanning
• Caching: Smart LRU cache for performance

Supported Languages (40+)

---

Setup & Development Guide

Prerequisites

• Node.js 18+
• pnpm (npm install -g pnpm)
• Rust (for native modules): rustup.rs
• C++ Build Tools:
  • macOS: xcode-select --install
  • Windows: Visual Studio Build Tools (C++ workload)
  • Linux: build-essential and cmake

Quick Setup

Automated

./scripts/setup.sh

Manual

# Install dependencies
pnpm install

# Build everything
pnpm run build

# Run extension (Press F5 in VS Code)

Development

# Watch mode (auto-recompile)
pnpm run watch

# Lint code
pnpm run lint

# Type check
pnpm run check-types

Testing

# All tests
pnpm test

# Unit tests only
pnpm run test:unit

# E2E tests only
pnpm run test:e2e

Troubleshooting

Rust not found

rustc --version
# If missing: curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

C++ build errors

• macOS: xcode-select --install
• Windows: Install Visual Studio Build Tools with C++ workload
• Linux: sudo apt-get install build-essential cmake

Module not found after build

pnpm install
pnpm run build

Extension won't activate

1. Check VS Code version (1.85.0+)
2. View logs: Ctrl+Shift+P → "Inline: Show Logs"
3. Rebuild: pnpm run build

---

🦾 Supported Models

Tier	VRAM	Recommended Models
Lightweight	2-4 GB	CodeGemma-2B, StableCode-3B, Phi-3-mini, TinyLlama-1.1B, Qwen1.5-0.5B-Chat
Mid-Tier	6-8 GB	DeepSeek-Coder-6.7B, StarCoder2-7B, CodeLlama-7B, WizardCoder-Python-7B, Phind-CodeLlama-34B (Q4_0)
Heavy	12GB+	CodeLlama-13B, Mixtral (Quantized), Dolphin-Mixtral-8x7B (Quantized), CodeLlama-34B
Ultra	24GB+	CodeLlama-34B, Llama-3-70B (Quantized), Yi-34B-200K, StarCoder2-15B

All models use GGUF format via llama.cpp.

---

🛠️ Commands

Press Ctrl+Shift+P and search for Inline:

• Model Manager - Manage and download models
• Toggle Offline Mode - Switch offline/online
• Clear Cache - Free memory
• Download Model from URL - Download from Hugging Face
• Show Logs - View debug information

---

Documentation

• Folder Structure: Explanation of the codebase layout.
• Contributing: Guidelines for submitting PRs.
• Testing: How to run the comprehensive test suite.

📄 License

This project is licensed under the Apache 2.0 License - see the LICENSE file for details.