TestGen AI is an LLM-based test generation tool that generates Dart test cases for uncovered code using Google Gemini to improve code coverage.

Features

• Coverage-Driven Test Generation: Automatically identifies untested code lines and generates tests to improve coverage.
• Dependency-Aware Context: Builds a dependency graph across code declarations by analyzing code dependencies to create dependency-aware context for prompting when testing any declaration. See how dependencies are included in prompts in the example/prompt_example.md.
• LLM Integration: Uses Google's Gemini models (Pro, Flash, Flash-Lite) for automated test generation with context-aware prompting.
• Iterative Validation: Validates generated tests through static analysis, execution, formatting, and optional coverage improvement checks with backoff propagation for API errors and rate limits.
• Smart Filtering: Skips trivial code (getters/setters, simple constructors) that doesn't require testing.

Getting Started

Install test_gen_ai

dart pub global activate test_gen_ai

Gemini API key

Running the package requires a Gemini API key.

• Configure your key using either method:
  • Set as environment variable:

    export GEMINI_API_KEY=your_api_key
    

  • Pass as command-line argument: --api-key your_api_key
• Obtain an API key at https://ai.google.dev/gemini-api/docs/api-key.

Usage

Generate tests for your entire package.

By default, this script assumes it's being run from the root directory of a package, and outputs test files to the test/testgen/ folder with the naming convention: {declaration_name}_{declaration_id}_{num_uncovered_lines}_test.dart

dart pub global run test_gen_ai:testgen

Advanced usage with custom configuration

dart pub global run test_gen_ai:testgen --package '/home/user/code' --model gemini-3-flash-preview --api-key your_key --max-depth 5 --max-attempts 10 --effective-tests-only -v

It’s recommended to run the package on a specific set of files rather than the entire codebase using target-files flag, This reduces execution time, and make results easier to analyze & review.

dart pub global run test_gen_ai:testgen --package '/home/user/code' --target-files 'lib/src/foo.dart,lib/src/temp.dart'

Command Line Options

Option	Short	Default	Description
--package		. (current directory)	Root directory of the package to test
--target-files		[]	Limit test generation to specific dart files inside the package (paths relative to package root, e.g. lib/foo.dart)
--model		gemini-3-flash-preview	Gemini model to use (gemini-3-flash-preview, gemini-2.5-flash, gemini-2.5-flash-lite)
--api-key		$GEMINI_API_KEY	Gemini API key for authentication
--effective-tests-only	-e	false	Only generate tests that actually improve coverage
--scope-output		[]	Restrict coverage to specific package paths
--max-depth		10	Maximum dependency depth for context generation
--max-attempts		5	Maximum number of attempts for test generation per declaration
--verbose	-v	false	Enable verbose logging (logs LLM prompts to a file)
--help	-h		Show usage information

⏰ Fair Warning

TestGen AI takes time - sometimes a lot of it. Depending on your codebase size, this might be a perfect time to:

• Grab a coffee ☕
• Take a power nap 😴
• Learn a new language 🗣️ (we recommend Dart!)
• Question your life choices that led to having so much untested code 🤔

The good news? You'll come back to beautifully generated tests.