AIRGAP StudioAIRGAP Studio

Getting Started

AIRGAP Studio installation and first launch guide

Getting Started

AIRGAP Studio is an offline AI IDE that delivers AI coding capabilities without an internet connection. This guide walks first-time users through installation and initial setup step by step.

Step 1: Download the Installer

Choose one of the following from the download page:

  • Standard Bundle (recommended, ~4.9 GB) — IDE + AI engine (llama-server) + extensions + Qwen3 8B model. Ready to use out of the box.
  • IDE Only — IDE + extensions only. Models are added separately via the Modelpack installer.
  • Modelpack — Standalone package of additional AI models (.gguf files).

The AI model files approach 5 GB in size, so the Standard Bundle is recommended.

Step 2: Install

Double-click the downloaded AIRGAPStudio-Standard-*.exe to launch the installer. Administrator privileges are not required. The default install location is %LOCALAPPDATA%\AIRGAP\Studio\.

Model files are automatically placed in build/vscodium/models/ (inside the install directory) and AIRGAP Studio auto-detects them on first launch.

Step 3: First Launch

Double-click the AIRGAP Studio icon on the desktop or launch it from the Start menu.

Startup sequence (automatic):

  1. AirgapLauncher (background) — Manages child process lifecycle via Windows Job Object
  2. llama-server (port 11434) — Vulkan GPU-accelerated AI engine
  3. Compatibility Proxy (port 11433) — Model compatibility gateway
  4. VSCodium — IDE shell (custom build)
  5. AIRGAP Lite Assistant (active by default) — Inline autocomplete + RAG chat

The first launch may take 1-2 minutes while the model loads.

Step 4: First Use — Chat with the Lite Assistant

Click the AIRGAP icon in the VSCodium Activity Bar (the vertical icon column on the left). A chat panel opens in the side panel.

  • Shortcuts (Continue fork base): Ctrl+L (Quick Edit), Ctrl+I (Inline Edit), Ctrl+Alt+Space (Autocomplete)
  • Type a question in English or Korean — the local Qwen3 8B model responds.

Step 5: (Optional) Switch to a Different AI Model

Open the Command Palette (Ctrl+Shift+P) and run AIRGAP: Select LLM Model. Choose from 6 supported models:

  • Qwen3 8B (default, recommended)
  • Qwen3 4B (for 4 GB VRAM environments)
  • Granite 4.0 Micro / Qwen3 1.7B (lightweight)
  • Mistral 7B Instruct v0.3 / Meta Llama 3.1 8B Instruct

After selection, llama-server automatically restarts and every extension's model display refreshes. (Monitor is the sole authority — see the Monitor docs for details.)

Step 6: (Optional) Switch to the Cline-based AIRGAP Assistant

If you need autonomous agent capabilities and multi-file refactoring:

Command Palette → AIRGAP: Switch Assistant → VSCodium restarts → AIRGAP Assistant (Cline 3.56.2 fork) activates.

The two assistants are mutually exclusive — only one is active at a time. See the Switching Assistant docs for details.

System Requirements Summary

  • OS: Windows 10 64-bit (1903+) or Windows 11
  • RAM: 16 GB minimum (32 GB recommended)
  • GPU: Vulkan-capable GPU recommended (NVIDIA / AMD / Intel Arc), 6 GB+ VRAM recommended. Below 4 GB falls back automatically to CPU mode (slower response times).
  • Disk: 10 GB or more (20 GB SSD recommended)

For full specifications, see the System Requirements document.

Next Steps