Release Notes

v1.1.2 (planned 2026-05-19) — UX Polish

A patch focused on tightening user-facing content accuracy and expanding documentation and FAQs.

Content Accuracy Cleanup

• Unified LLM engine references to llama-server (cleaned up leftover Ollama mentions)
• Removed references to non-existent models and updated the supported model list

New Documentation

• AIRGAP Lite Assistant guide
• AIRGAP Monitor guide
• AIRGAP: Switch Assistant usage notes
• Supported Models (list of bundled models)
• Compatibility Proxy overview
• LLM Server Troubleshooting

FAQ

• Added 10 frequently asked questions

---

v1.1.1 (2026-05-12)

A minor release focused on stability improvements and automation against defect accumulation.

AIRGAP Assistant (Cline fork) restored

• The Liberation cycle restored the Cline fork bundle to a healthy state
• dist/extension.js (~19.5 MB) and webview-ui/ build artifacts are now produced correctly
• The chat webview works correctly when selected via Switch Assistant

Launcher 10-Layer prevention gate

• Regression-prevention gates are now executed automatically during installer builds
• Validates llama-server argument compatibility, manifest ↔ binary alignment, dev/prod emit-set diffs, dry-run health, and more — layer by layer
• Strengthens automation to prevent defect accumulation

---

v1.1.0 (2026-04-21) — Multi-Model + Dual Assistant

A major release introducing multi-model support and a dual-assistant architecture.

LLM Engine Replacement

• Switched from Ollama to llama-server (Vulkan GPU)
• Supports NVIDIA, AMD, and Intel Arc GPUs

Multi-Model Support

• Six bundled models: Qwen3 1.7B / 4B / 8B, Granite 4.0 Micro, Mistral 7B Instruct v0.3, Meta Llama 3.1 8B Instruct
• Default model: Qwen3 8B
• .gguf files placed under build/vscodium/models/ are auto-discovered

New: AIRGAP Lite Assistant

• Based on a Continue 1.3.39 fork
• Inline autocomplete + RAG + Quick Edit
• Active by default on fresh installs

New: AIRGAP Monitor

• An extension solely responsible for model selection
• LLM status monitoring UI

New Command: AIRGAP: Switch Assistant

• Switches between the two assistants (Cline fork ↔ Continue fork)
• Applied by restarting VSCodium

New: Compatibility Proxy (port 11433)

• A chat-format normalization gateway between LLM-consuming extensions and llama-server
• Per-model / per-client profile routing
• The dev launcher automatically patches the endpoint

Installer Split into 3 Variants

• Full
• IDE-Only
• Model Pack

---

v1.0.0 (Initial Release)

The first official release of AIRGAP Studio. Provides a complete AI development environment for air-gapped networks.

IDE Foundation

• Custom-built IDE
• Windows 10/11 64-bit support
• Offline-only architecture (telemetry disabled)

Bundled Extensions (7 user-facing + 4 infrastructure)

User-facing extensions (7):

Additional infrastructure extensions (monitor, launcher, runner, shield) are bundled as well. AIRGAP Assistant and AIRGAP Lite Assistant are mutually exclusive — only one is active at a time.

AI Model

• Default model: Qwen3 8B (local AI engine, llama-server with Vulkan backend)
• Native tool calling support
• All LLM-connected extensions share the same model instance, requiring no additional VRAM
• AIRGAP Assistant supports four providers — openai, openai-native, openai-codex, and lmstudio — defaulting to the local llama-server. AIRGAP Lite Assistant uses llama-server exclusively by default; external foundation providers (Anthropic, OpenAI, Google Gemini) can be enabled via opt-in configuration.

Key Features

• AIRGAP Assistant: AI code generation, refactoring, and debugging support
• AIRGAP Designer: Natural language-based UI prototyping with HTML/CSS generation
• AIRGAP Bridge: Extracts design tokens from Figma design files and converts them to code
• AIRGAP UI Library: Preview, copy code, and insert 85 components directly into the editor
• AIRGAP Autocomplete: Real-time code suggestions while typing (300ms debounce)
• AIRGAP License: Offline license authentication using ED25519 signatures

Known Limitations

• 7 web components (<uk-*>) do not render in the preview due to unbundled JS.
• Figma MCP features require the Figma Desktop app and MCP plugin.
• The code conversion engine runs as an independent process.