AI Test Reporter is a powerful tool that generates intelligent summaries of test results using a wide range of AI models. With access to over 300 models through various providers (OpenAI, Anthropic Claude, Google Gemini, Mistral, Perplexity, OpenRouter, and more), it analyzes failing tests and provides actionable insights about what went wrong and how to fix it.
AI Test Reporter is community-built and open source.
You can support the project by:
- Giving this repository a ⭐
- Following the @ctrf organization on GitHub
- 🤖 Smart Analysis: Get AI-powered explanations of why tests failed and suggestions for fixes
- 🔌 Multiple Providers: Choose from 300+ AI models across different providers
- 🛠️ Custom Services: Use your own OpenAI-compatible services (in-house solutions, LocalAI, Ollama, etc.)
- 💰 Cost-Effective Options: Start with providers offering free credits like Mistral and Google Gemini
- 🔄 Developer Tool Integration: Seamlessly integrate AI summaries into your workflow:
- GitHub Actions and Pull Requests
- Slack / Teams notifications
- And more through standardized CTRF reports
- 🎯 Consolidated Insights: Get high-level summaries of test suite failures to identify patterns and root causes
- 📊 Structured JSON Analysis: Generate categorized breakdown of issues (code, timeout, application) with actionable recommendations
- ⚡ Framework Agnostic: Works with any testing framework through CTRF reports
- 🔗 Flexible Reporting: AI summaries are embedded in your CTRF JSON report, allowing you to:
- Include AI analysis in your existing reporting workflows
- Programmatically customize how and where summaries appear in your Developer Tools
- Export structured JSON for integration with monitoring and analytics tools
Tip
The following integrations are available with free tiers so are great to try out AI test reporter:
- Google Gemini: Offers a free tier with lower rate limits. Pricing details
- Mistral AI: Provides a free API tier to explore the service. Tier details
You can use any of the models supported by the following providers:
- OpenAI
- Anthropic Claude
- Google Gemini
- Mistral
- Grok
- DeepSeek
- Azure OpenAI
- Perplexity
- OpenRouter
- Custom OpenAI-compatible services (in-house solutions, LocalAI, Ollama, etc.)
You use your own API keys for the models you select.
Generate a CTRF report using your testing framework. CTRF reporters are available for most testing frameworks and easy to install.
No CTRF reporter? No problem!
Use junit-to-ctrf to convert a JUnit report to CTRF
Run the following command:
npx ai-ctrf openai <path-to-ctrf-report>An AI summary for each failed test will be added to your test report.
The package interacts with the OpenAI API, you must set OPENAI_API_KEY environment variable.
You will be responsible for any charges incurred from using your selected OpenAI model. Make sure you are aware of the associated cost.
A message is sent to OpenAI for each failed test.
--model: OpenAI model to use (default: gpt-3.5-turbo).
--systemPrompt: Custom system prompt to guide the AI response.
--additionalSystemPromptContext: Additional context to append to the system prompt. This allows you to add extra instructions or context without replacing the entire system prompt.
--additionalPromptContext: Additional context to append to the user prompt. This allows you to add extra instructions
--frequencyPenalty: OpenAI frequency penalty parameter (default: 0).
--maxTokens: Maximum number of tokens for the response.
--presencePenalty: OpenAI presence penalty parameter (default: 0).
--temperature: Sampling temperature (conflicts with topP).
--topP: Top-p sampling parameter (conflicts with temperature).
--log: Whether to log the AI responses to the console (default: true).
--maxMessages: Limit the number of failing tests to send for summarization in the LLM request. This helps avoid overwhelming the model when dealing with reports that have many failing tests. (default: 10)
--consolidate: Consolidate and summarize multiple AI summaries into a higher-level overview (default: true)
--json-analysis: Generate structured JSON analysis with categorized issues (code, timeout, application) and recommendations (default: false)
Run the following command:
npx ai-ctrf azure-openai <path-to-ctrf-report>An AI summary for each failed test will be added to your test report.
The package interacts with the Azure OpenAI API, you must set AZURE_OPENAI_API_KEY, AZURE_OPENAI_ENDPOINT, and AZURE_OPENAI_DEPLOYMENT_NAME environment variable or provide them as arguments.
You will be responsible for any charges incurred from using your selected Azure OpenAI model. Make sure you are aware of the associated cost.
A message is sent to Azure OpenAI for each failed test.
--model: OpenAI model to use (default: gpt-3.5-turbo).
--systemPrompt: Custom system prompt to guide the AI response.
--additionalSystemPromptContext: Additional context to append to the system prompt. This allows you to add extra instructions or context without replacing the entire system prompt.
--additionalPromptContext: Additional context to append to the user prompt.
--frequencyPenalty: OpenAI frequency penalty parameter (default: 0).
--maxTokens: Maximum number of tokens for the response.
--presencePenalty: OpenAI presence penalty parameter (default: 0).
--temperature: Sampling temperature (conflicts with topP).
--topP: Top-p sampling parameter (conflicts with temperature).
--log: Whether to log the AI responses to the console (default: true).
--maxMessages: Limit the number of failing tests to send for summarization in the LLM request. This helps avoid overwhelming the model when dealing with reports that have many failing tests. (default: 10)
--consolidate: Consolidate and summarize multiple AI summaries into a higher-level overview (default: true)
--json-analysis: Generate structured JSON analysis with categorized issues (code, timeout, application) and recommendations (default: false)
Run the following command:
npx ai-ctrf claude <path-to-ctrf-report>An AI summary for each failed test will be added to your test report.
The package interacts with the Anthropic API, you must set ANTHROPIC_API_KEY environment variable.
You will be responsible for any charges incurred from using your selected Claude model. Make sure you are aware of the associated costs.
A message is sent to Claude for each failed test.
--model: Claude model to use (default: claude-3-5-sonnet-20240620).
--systemPrompt: Custom system prompt to guide the AI response.
--additionalSystemPromptContext: Additional context to append to the system prompt.
--additionalPromptContext: Additional context to append to the user prompt.
--maxTokens: Maximum number of tokens for the response.
--temperature: Sampling temperature.
--log: Whether to log the AI responses to the console (default: true).
--maxMessages: Limit the number of failing tests to send for summarization in the LLM request. This helps avoid overwhelming the model when dealing with reports that have many failing tests. (default: 10)
--consolidate: Consolidate and summarize multiple AI summaries into a higher-level overview (default: true)
--json-analysis: Generate structured JSON analysis with categorized issues (code, timeout, application) and recommendations (default: false)
Run the following command:
npx ai-ctrf grok <path-to-ctrf-report>An AI summary for each failed test will be added to your test report.
The package interacts with the Grok API, you must set GROK_API_KEY environment variable.
You will be responsible for any charges incurred from using your selected Grok model. Make sure you are aware of the associated cost.
A message is sent to Grok for each failed test.
--model: Grok model to use (default: grok-1).
--systemPrompt: Custom system prompt to guide the AI response.
--additionalSystemPromptContext: Additional context to append to the system prompt.
--additionalPromptContext: Additional context to append to the user prompt.
--maxTokens: Maximum number of tokens for the response.
--temperature: Sampling temperature.
--log: Whether to log the AI responses to the console (default: true).
--maxMessages: Limit the number of failing tests to send for summarization in the LLM request. This helps avoid overwhelming the model when dealing with reports that have many failing tests. (default: 10)
--consolidate: Consolidate and summarize multiple AI summaries into a higher-level overview (default: true)
--json-analysis: Generate structured JSON analysis with categorized issues (code, timeout, application) and recommendations (default: false)
Run the following command:
npx ai-ctrf deepseek <path-to-ctrf-report>An AI summary for each failed test will be added to your test report.
The package interacts with the DeepSeek API, you must set DEEPSEEK_API_KEY environment variable.
You will be responsible for any charges incurred from using your selected DeepSeek model. Make sure you are aware of the associated cost.
A message is sent to DeepSeek for each failed test.
--model: DeepSeek model to use (default: deepseek-coder).
--systemPrompt: Custom system prompt to guide the AI response.
--additionalSystemPromptContext: Additional context to append to the system prompt.
--additionalPromptContext: Additional context to append to the user prompt.
--maxTokens: Maximum number of tokens for the response.
--temperature: Sampling temperature.
--log: Whether to log the AI responses to the console (default: true).
--maxMessages: Limit the number of failing tests to send for summarization in the LLM request. This helps avoid overwhelming the model when dealing with reports that have many failing tests. (default: 10)
--consolidate: Consolidate and summarize multiple AI summaries into a higher-level overview (default: true)
--json-analysis: Generate structured JSON analysis with categorized issues (code, timeout, application) and recommendations (default: false)
Run the following command:
npx ai-ctrf mistral <path-to-ctrf-report>An AI summary for each failed test will be added to your test report.
The package interacts with the Mistral API, you must set MISTRAL_API_KEY environment variable.
You will be responsible for any charges incurred from using your selected Mistral model. Make sure you are aware of the associated cost.
A message is sent to Mistral for each failed test.
Mistral offers free API credits upon signup, making it a great option to test the tool without immediate costs.
--model: Mistral model to use (default: mistral-medium).
--systemPrompt: Custom system prompt to guide the AI response.
--additionalSystemPromptContext: Additional context to append to the system prompt.
--additionalPromptContext: Additional context to append to the user prompt.
--maxTokens: Maximum number of tokens for the response.
--temperature: Sampling temperature.
--topP: Top-p sampling parameter.
--log: Whether to log the AI responses to the console (default: true).
--maxMessages: Limit the number of failing tests to send for summarization in the LLM request. This helps avoid overwhelming the model when dealing with reports that have many failing tests. (default: 10)
--consolidate: Consolidate and summarize multiple AI summaries into a higher-level overview (default: true)
--json-analysis: Generate structured JSON analysis with categorized issues (code, timeout, application) and recommendations (default: false)
Run the following command:
npx ai-ctrf gemini <path-to-ctrf-report>An AI summary for each failed test will be added to your test report.
The package interacts with the Google Gemini API, you must set GOOGLE_API_KEY environment variable.
You will be responsible for any charges incurred from using your selected Gemini model. Make sure you are aware of the associated cost.
A message is sent to Gemini for each failed test.
Google offers free API credits for Gemini, providing a cost-effective way to try out the tool.
--model: Gemini model to use (default: gemini-pro).
--systemPrompt: Custom system prompt to guide the AI response.
--additionalSystemPromptContext: Additional context to append to the system prompt.
--additionalPromptContext: Additional context to append to the user prompt.
--maxTokens: Maximum number of tokens for the response.
--temperature: Sampling temperature.
--topP: Top-p sampling parameter.
--log: Whether to log the AI responses to the console (default: true).
--maxMessages: Limit the number of failing tests to send for summarization in the LLM request. This helps avoid overwhelming the model when dealing with reports that have many failing tests. (default: 10)
--consolidate: Consolidate and summarize multiple AI summaries into a higher-level overview (default: true)
--json-analysis: Generate structured JSON analysis with categorized issues (code, timeout, application) and recommendations (default: false)
Run the following command:
npx ai-ctrf perplexity <path-to-ctrf-report>An AI summary for each failed test will be added to your test report.
The package interacts with the Perplexity API, you must set PERPLEXITY_API_KEY environment variable.
You will be responsible for any charges incurred from using your selected Perplexity model. Make sure you are aware of the associated cost.
A message is sent to Perplexity for each failed test.
--model: Perplexity model to use (default: pplx-7b-online).
--systemPrompt: Custom system prompt to guide the AI response.
--additionalSystemPromptContext: Additional context to append to the system prompt.
--additionalPromptContext: Additional context to append to the user prompt.
--maxTokens: Maximum number of tokens for the response.
--temperature: Sampling temperature.
--topP: Top-p sampling parameter.
--log: Whether to log the AI responses to the console (default: true).
--maxMessages: Limit the number of failing tests to send for summarization in the LLM request. This helps avoid overwhelming the model when dealing with reports that have many failing tests. (default: 10)
--consolidate: Consolidate and summarize multiple AI summaries into a higher-level overview (default: true)
--json-analysis: Generate structured JSON analysis with categorized issues (code, timeout, application) and recommendations (default: false)
Run the following command:
npx ai-ctrf openrouter <path-to-ctrf-report>An AI summary for each failed test will be added to your test report.
The package interacts with the OpenRouter API, you must set OPENROUTER_API_KEY environment variable.
OpenRouter provides access to multiple AI models through a single API, including models from Anthropic, OpenAI, Google, Meta, and more.
You will be responsible for any charges incurred from using your selected OpenRouter model. Make sure you are aware of the associated cost.
A message is sent to OpenRouter for each failed test.
--model: OpenRouter model to use (default: anthropic/claude-3-opus). Available models include:
- anthropic/claude-3-opus
- google/gemini-pro
- meta-llama/llama-2-70b-chat
- mistral/mixtral-8x7b And many more from the OpenRouter catalog.
--systemPrompt: Custom system prompt to guide the AI response.
--additionalSystemPromptContext: Additional context to append to the system prompt.
--additionalPromptContext: Additional context to append to the user prompt.
--maxTokens: Maximum number of tokens for the response.
--temperature: Sampling temperature.
--topP: Top-p sampling parameter.
--frequencyPenalty: Frequency penalty parameter.
--presencePenalty: Presence penalty parameter.
--log: Whether to log the AI responses to the console (default: true).
--maxMessages: Limit the number of failing tests to send for summarization in the LLM request. This helps avoid overwhelming the model when dealing with reports that have many failing tests. (default: 10)
--consolidate: Consolidate and summarize multiple AI summaries into a higher-level overview (default: true)
--json-analysis: Generate structured JSON analysis with categorized issues (code, timeout, application) and recommendations (default: false)
Run the following command:
npx ai-ctrf custom <path-to-ctrf-report> --url <your-custom-api-url>An AI summary for each failed test will be added to your test report.
This option allows you to use any OpenAI-compatible API endpoint. This is useful for:
- Self-hosted services (like LocalAI, text-generation-webui, etc.)
- Custom in-house AI solutions
- Local development with Ollama's OpenAI-compatible endpoint
- Other OpenAI-compatible services
You can set the custom URL via environment variable instead of the CLI option:
export AI_CTRF_CUSTOM_URL=http://localhost:8080/v1For authentication, you can use:
export AI_CTRF_CUSTOM_API_KEY=your-api-keyIf no API key is provided, the tool will fall back to checking OPENAI_API_KEY or use a placeholder value if the API doesn't require authentication.
--url: Base URL for the custom OpenAI-compatible API (e.g., http://localhost:8080/v1). Can also be set via AI_CTRF_CUSTOM_URL environment variable.
--model: Model to use (default: gpt-4o). This should match the model name your custom API expects.
--systemPrompt: Custom system prompt to guide the AI response.
--additionalSystemPromptContext: Additional context to append to the system prompt.
--additionalPromptContext: Additional context to append to the user prompt.
--frequencyPenalty: Frequency penalty parameter (default: 0).
--maxTokens: Maximum number of tokens for the response.
--presencePenalty: Presence penalty parameter (default: 0).
--temperature: Sampling temperature (conflicts with topP).
--topP: Top-p sampling parameter (conflicts with temperature).
--log: Whether to log the AI responses to the console (default: true).
--maxMessages: Limit the number of failing tests to send for summarization in the LLM request. This helps avoid overwhelming the model when dealing with reports that have many failing tests. (default: 10)
--consolidate: Consolidate and summarize multiple AI summaries into a higher-level overview (default: true)
--json-analysis: Generate structured JSON analysis with categorized issues (code, timeout, application) and recommendations (default: false)
Using Ollama with OpenAI-compatible endpoint:
# Start Ollama with OpenAI-compatible API
ollama serve
# Run AI test reporter
npx ai-ctrf custom ctrf-report.json --url http://localhost:11434/v1 --model llama2In addition to generating individual test summaries, AI Test Reporter can provide a structured JSON analysis that categorizes issues and provides actionable recommendations.
The --json-analysis flag generates a structured breakdown of your test failures:
{
"summary": "High-level overview of what went wrong",
"code_issues": "Detailed description of code-related problems",
"timeout_issues": "Detailed description of timeout and performance issues",
"application_issues": "Detailed description of application-level problems",
"recommendations": "Actionable recommendations to fix the issues"
}Add the --json-analysis flag to any model command:
# OpenAI
npx ai-ctrf openai ctrf-report.json --json-analysis
# Claude
npx ai-ctrf claude ctrf-report.json --json-analysis
# Any other model
npx ai-ctrf <model> ctrf-report.json --json-analysis# Parse specific fields in CI/CD
result=$(npx ai-ctrf openai report.json --json-analysis --log false)
echo "$result" | jq '.recommendations'
# Send to monitoring or analytics tools
npx ai-ctrf openai report.json --json-analysis --log false | \
curl -X POST https://api.example.com/test-analysis
# Save for later processing
npx ai-ctrf openai report.json --json-analysis > analysis.jsonThe --additionalSystemPromptContext and --additionalPromptContext options allow you to provide extra context to help the AI better understand and analyze your test failures. These are available for all AI providers.
# Add project-specific context
npx ai-ctrf openai ctrf-report.json \
--additionalPromptContext "This is from our production environment running on AWS" \
--additionalSystemPromptContext "Focus on database connection and timeout issues"
# Add team-specific guidelines
npx ai-ctrf claude ctrf-report.json \
--additionalSystemPromptContext "Our team uses React 18 and follows the Airbnb style guide" \
--additionalPromptContext "Tests are running in CI/CD pipeline on GitHub Actions"The AI model analyzes information to:
- Understand the complete context of the failure
- Identify potential root causes
- Suggest specific fixes
- Highlight patterns across multiple failures
When consolidation is enabled (--consolidate), the AI analyzes all test failures AI summaries together to provide a high-level summary of issues.
For each failing test, the AI receives the complete test object.
The complete environment details from the environment object in the CTRF report is provided to the AI model.
The complete tool details from the tool object in the CTRF report is provided to the AI model.
You are responsible for any charges incurred from using the AI models. Make sure you are aware of the associated costs.
{
"results": {
"tool": {
"name": "AnyFramework"
},
"summary": {
"tests": 1,
"passed": 0,
"failed": 1,
"pending": 0,
"skipped": 0,
"other": 1,
"start": 1722511783500,
"stop": 1722511804528
},
"tests": [
{
"name": "should display profile information",
"status": "failed",
"duration": 800,
"message": "Assertion Failure: profile mismatch",
"trace": "ProfileTest.js:45...",
"ai": "The test failed because there was a profile mismatch at line 45 of the ProfileTest.js file. To resolve this issue, review the code at line 45 to ensure that the expected profile information matches the actual data being displayed. Check for any discrepancies and make necessary adjustments to align the expected and actual profile information."
}
]
}
}
View AI summaries in directly in the Github Actions workflow:
Add a Pull Request comment with your AI summary:
Send a Slack message with your AI test summary:
If you find this project useful, consider giving it a GitHub star ⭐ It means a lot to us.
