GPT Driver User Guide
GPT Driver User Guide
GPT Driver User Guide
  • Getting Started
    • πŸš€Getting Started
    • πŸ”ƒUploading a Build File
    • πŸ§™β€β™‚οΈCreating Tests
      • Commands
        • Tap Command
        • Type Command
        • scroll Command
        • scrollUntilVisible Command
        • swipe Command
        • slide Command
        • wait Command
        • assertVisible Command
        • openLink Command
        • pressBackButton Command
        • launchApp Command
        • restartApp Command
      • πŸ‘οΈwithVision: Instructions
      • 🧠AI Instructions
    • 🏁Executing Tests
  • ☎️Device Configuration
  • βš™οΈUnder the Hood
    • Test Generation
    • Test Execution
  • Getting Around
    • ✏️Test Editor
    • πŸ›°οΈTest Overview
    • πŸ…Test Reports
    • ⏺️Test Recordings
    • πŸ‘€Settings
  • Best Practices
    • πŸ§‘β€πŸ’»API Documentation
    • Versioning
    • ↗️Templates
    • πŸ–‡οΈTest Dependencies & Prompt References
    • πŸ”—Deep Links
    • πŸ“§Email Verification
    • πŸ“‘Network Calls
    • πŸͺ‘Parameterized Strings
    • πŸ“Changing Device Location
    • πŸͺΆConditional Steps
    • 🐦Nested Steps
    • ⌚Smart Caching
    • πŸ—£οΈEnv. Variables
    • πŸ’―Bulk Step Testing for Robustness
    • πŸ“–Exact Text Assertions
    • πŸ’¬Auto-grant Permissions
  • πŸ§ͺMocking Network Data
  • 🌎Localization Testing
  • Code Generation
  • ❔FAQs
Powered by GitBook
On this page
  • Under the Hood
  • Covered Topics
  • When To Use This

Under the Hood

This section explains the inner workings of our test generation and automation engine for engineers and advanced users, covering its logic, safeguards, and design choices.

Under the Hood

This section documents the internals of GPT Driver's automation engine. It’s intended for engineers debugging edge cases, reviewing system guarantees, or integrating with downstream tools.

Covered Topics

  • Test Generation Converts plain-language inputs into deterministic, idempotent test cases using a fine-tuned Llama-3 70B model. Includes tagging, escape hatches, prompt templates, and drift control via test case hashing.

  • Test Execution Executes test cases using a layered resolution stack: command parsing, UI element detection, and AI fallback logic. Includes scroll behavior, popup handling, screen stability checks, and fail-safe conditions.

When To Use This

  • Need to validate hash consistency or trace why a test didn’t regenerate.

  • Diagnosing AI-based interaction fallbacks during execution.

  • Reviewing how UI stability or infinite loop detection is enforced.

Each subpage focuses on logic and safeguards, not UI-level workflows.

PreviousDevice ConfigurationNextTest Generation

Last updated 4 days ago

βš™οΈ