App Blink integrates frontier language models directly into your IDE without requiring separate OpenAI, Anthropic, or xAI API keys. By plugging into your existing GitHub Copilot subscription or the Antigravity CLI (agy), App Blink eliminates third-party subscription overhead.


🧠 Supported Frontier Models

You can switch models on the fly through VS Code settings (appblink.ai.models.textGeneration) or via the command palette:

Model Provider Context Window Best Suited For
Claude Sonnet 4.6 Anthropic 160K Balanced reasoning, refactoring, code generation
Claude Opus 4.6 Anthropic 192K Deep architectural reasoning, multi-file migrations
Claude Haiku 4.5 Anthropic 160K Ultra-fast tasks, real-time translations, unit tests
GPT-5.4 OpenAI 400K High-intelligence code generation & broad reasoning
GPT-5.3-Codex OpenAI 400K Code-optimized logic, algorithm optimization
Gemini 3.6 Flash Google 1M+ Rapid multimodal analysis, massive context processing
Gemini 3.5 Flash Google 1M+ High-speed generalist model
Gemini 3.1 Flash-Lite Google 1M+ Sub-second translations on save (app_en.arb)
Grok Code Fast 1 xAI 173K Fast tool execution & code transformation

⚑ Multi-Tier AI Provider Resolution

When an App Blink action requires AI generation (store metadata, test suites, marketing copy, or VibeCheck reviews), the routing abstraction resolves providers in the following order:

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚  PROVIDER RESOLUTION PIPELINE                          β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚ 1. Antigravity CLI (`agy`)                             β”‚
β”‚    Uses active system credentials with zero config.    β”‚
β”‚                                                        β”‚
β”‚ 2. GitHub Copilot Language Model API (`vscode.lm`)     β”‚
β”‚    Uses active GitHub Copilot entitlement.             β”‚
β”‚                                                        β”‚
β”‚ 3. Direct Google Gemini API Key                        β”‚
β”‚    Retrieved securely from OS `SecretStorage`.         β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

πŸ“Š Real-Time Token Tracker & Cost Analysis

The Usage panel in the App Blink sidebar continuously tracks LLM usage:

  • Session Token Monitor: Displays prompt tokens, completion tokens, and total usage for your active editing session.
  • Cost Estimation: Compares your Copilot/Antigravity token usage against commercial SaaS API rates (e.g. standard Anthropic/OpenAI token costs) to show cumulative dollar savings.
  • Clear Usage: Reset counters anytime using App Blink: Clear Usage Statistics (appblink.clearUsageStats).

βš™οΈ Configuration Reference (`settings.json`)

{
  // Primary text model for metadata, release notes, and translations
  "appblink.ai.models.textGeneration": "antigravity-cli",

  // Model forwarded when using the Antigravity backend
  "appblink.ai.antigravity.model": "claude-sonnet-4-6",

  // Maximum runtime in minutes before an agent task times out
  "appblink.ai.antigravity.maxMinutes": 10,

  // AI model used for logo and asset generation
  "appblink.ai.models.imageGeneration": "gemini-3.1-flash-lite-image",

  // AI model used for promo video generation
  "appblink.ai.models.videoGeneration": "veo-3.1-lite-generate-preview"
}