AIRGAP StudioSystem Requirements
System requirements for installing AIRGAP Studio
System Requirements
AIRGAP Studio is an offline AI IDE that runs without an internet connection. This page summarizes the hardware and software requirements for stable operation.
Operating System
| Item | Minimum | Recommended |
|---|---|---|
| Windows | Windows 10 64-bit (build 1903 or later) | Windows 11 |
macOS / Linux are not currently supported.
Memory (RAM)
| Item | Minimum | Recommended |
|---|---|---|
| RAM | 16 GB | 32 GB |
The IDE, AI engine (llama-server), and model run concurrently, so less than 16 GB is not recommended.
GPU
The AIRGAP Studio AI engine uses the Vulkan backend. All Vulkan-capable GPUs (NVIDIA, AMD, Intel Arc) are compatible.
| Item | Minimum | Recommended |
|---|---|---|
| GPU | Vulkan-capable GPU (under 4 GB VRAM auto-falls back to CPU) | Vulkan-capable GPU with 6 GB+ VRAM (smooth Qwen3 8B) |
No CUDA driver required. Only a Vulkan driver (included in Windows Update or your GPU vendor driver) is needed.
Less Than 4 GB VRAM — Automatic CPU Fallback
The Launcher's GpuFallback mechanism detects VRAM under 4096 MB and automatically switches to CPU inference mode. AI responses slow down but remain functional.
When using lightweight models (Qwen3 1.7B, Granite 4.0 Micro), GPU acceleration is possible even on 4 GB environments.
Disk Space
| Item | Size |
|---|---|
| Standard bundle installer | ~4.9 GB |
| After installation (IDE + llama-server + Qwen3 8B model) | ~7 GB |
| Recommended free space | 10 GB or more (20 GB SSD recommended) |
Required / Optional Software
| Item | Necessity | Notes |
|---|---|---|
| llama-server (AI engine, Vulkan) | Required | Bundled in Standard installer — no separate install needed |
| Qwen3 8B model (.gguf) | Required | Bundled in Standard installer. Additional models via the Modelpack installer |
| VSCodium 1.109.x | Required | Bundled in installer (VSCodium open-source build, not VS Code) |
| Node.js 18+ (24.x recommended) | Optional | For running MCP servers. Not required for AIRGAP Studio itself |
| Vulkan driver | Required | NVIDIA / AMD / Intel Arc all work. Independent of CUDA |
Ollama, required in older versions, is no longer needed (replaced by llama-server starting in v1.1.0).
Network
The core features of AIRGAP Studio (local AI coding assistant, inline autocomplete, RAG chat) require no internet connection.
The following features require an internet connection (optional):
- Foundation Providers (OpenAI, Anthropic, Gemini, etc.) — opt-in feature of AIRGAP Lite Assistant
- External remote MCP server connections
- Figma MCP (localhost:3845, when using Figma Desktop)
Port Usage
| Port | Purpose | Configuration |
|---|---|---|
| 11434 | llama-server (AI engine) | Auto-start |
| 11433 | Compatibility Proxy | Auto-start |
| 9222 | Chrome DevTools (dev mode) | Only when using dev-launcher.ps1 |
Diagnostics
Verify proper operation after installation:
- Launch AIRGAP Studio
- In PowerShell:
Invoke-WebRequest http://127.0.0.1:11434/health -UseBasicParsing→ HTTP 200 - Click the AIRGAP icon in the Activity Bar → confirm the Lite Assistant chat panel appears
If you encounter issues, see LLM Server Troubleshooting or Compatibility Proxy.