Tips & Tricks

Using the AI Assistant

Give Clear Instructions

Being specific with your requests to the AI assistant yields more accurate results.

# Inefficient request
"Fix this code"

# Effective request
"Add a null check to the handleSubmit function and show a toast notification to the user when an error occurs"

Provide Context

The AI can better understand context when you have files open or code selected. If multiple files are relevant, reference them with @filename.

Use Auto Compact

Long conversations can exhaust the context window. AIRGAP Assistant has Auto Compact enabled by default, which automatically summarizes the conversation.

Design Workflow

Bridge to Designer Integration

1. Generate design tokens in AIRGAP Bridge.
2. The tokens are automatically injected into AIRGAP Designer's system prompt.
3. When generating UI in Designer, you get consistent output with your design tokens applied.

Using UI Library Components

Search for components in the UI Library and use the Copy Code or Insert into Editor features.

Performance Optimization

GPU Acceleration

If you have an NVIDIA, AMD, or Intel GPU, the local AI engine (llama-server, Vulkan backend) will automatically use it. You can verify GPU usage via the Launcher UI status panel, the health endpoint at http://127.0.0.1:11434/health, or by running AIRGAP: Refresh Status from the Command Palette.

Adjusting Context Size

If AI responses are incomplete for complex tasks, check the context size. The default is 16,384 tokens and can be adjusted in the settings.

Disable Unused Extensions

Disabling unused extensions speeds up IDE startup. Select Extensions: Show Installed Extensions from the Command Palette (Ctrl+Shift+P).

Layout Setup

Recommended Dual Monitor Layout

Recommended Single Monitor Layout

• Place the editor on the left and AIRGAP Assistant in the right sidebar.
• Toggle the file explorer with Ctrl+B to free up screen space.
• Show the bottom panel (terminal) only when needed with Ctrl+J.

Changing Models

To switch models, use the AIRGAP: Select LLM Model command from the Command Palette — this is handled exclusively by the AIRGAP Monitor extension. AIRGAP Assistant (Cline fork) supports four provider types: openai, openai-native, openai-codex, and lmstudio. AIRGAP Lite Assistant (Continue fork) operates with llama-server by default. External foundation providers (Anthropic, OpenAI, Google Gemini) can be enabled via opt-in settings in network-connected environments. See Supported Models for the full model catalog.