I released the first version of Rentgen on November 9, 2025. On October 28, 2026, we released v2.0.0.
The original workflow remains simple:
Paste a cURL → map the fields → run API tests.
Rentgen generates and executes deterministic checks for input validation, boundaries, authentication, security, and performance. That core workflow runs locally, is free, and requires no AI model.
Version 2 adds AI analysis and additional test generation, OpenAPI import, and a more compact interface.
Why add AI to an existing test engine?
After Rentgen finishes a run, it already has useful context:
- The original request.
- Field types, required fields, allowed values, and boundaries supplied through mapping.
- Requests that were actually executed.
- Actual responses and measurements.
- Configured expectations and test results.
That gives a model a specific task: analyse the recorded results and identify meaningful gaps in coverage.
For example, testing a required field with null does not establish what happens when the field is omitted entirely. A server error from a request that changed several fields may warrant a new test that isolates one condition.
The goal is to find useful next steps based on what has already happened.
Deeper AI Testing
Rentgen v2 supports connections to:
- OpenAI.
- Anthropic.
- Ollama, including models running locally.
You configure your provider and model connection. Rentgen prepares the analysis request automatically, so there is no prompt to write.
The analysis returns:
- A summary of observed API behaviour and coverage.
- Up to three important findings, with evidence and impact.
- Areas that need further investigation.
- Up to ten additional executable tests covering meaningful gaps.
Ten is a limit, not a target. If only two useful tests remain, generating eight more variations adds noise.
You can also provide additional API context, such as business rules or areas you want investigated.
Generated tests become part of your project
The output is structured test data that Rentgen can use directly.
You can inspect each proposed request, review its expected result, and execute it in Rentgen.
AI helps discover a test, then the test runner executes and checks request deterministically.
The model’s suggestion still needs review. It can misunderstand a business rule or propose an incorrect expectation. Rentgen lets you inspect and adjust the test before making it part of your ongoing coverage.
Your model connection
AI requests go directly from your device to the provider you configure.
Rentgen has no backend receiving your API requests, reports, or model credentials. Credentials are stored locally.
If you connect a local Ollama model, analysis can stay on your own machine. If you choose a cloud provider, the supplied context goes to that provider.
OpenAPI import
You can now import an OpenAPI JSON file or provide its URL.
Rentgen creates a folder and collection of endpoint requests from the specification. This gives you a starting point for configuring and testing an API without importing each request individually.
You still supply the environment, credentials, and any values needed to make those requests work.
A more compact interface
Version 2 reorganises test results into tabs:
- Deeper AI Testing.
- Security Tests.
- Performance Insights.
- Data-Driven Tests.
- CRUD.
This makes larger runs easier to navigate and keeps each suite’s results together.
The release also includes the QUERY method, GUID field mapping, expanded numeric ranges, fixes for mapping persistence after errors, and other usability and stability improvements.
What remains free?
The deterministic API testing engine remains free.
Deeper AI Testing is an optional Rentgen Pro feature, priced at $36 per year. You bring your own model connection; cloud model usage is billed separately by your provider.
Try Rentgen v2
Download Rentgen, import a cURL or OpenAPI specification, configure the fields, and run the deterministic tests.
If you try the AI integration, I’m particularly interested in whether its additional tests find something useful beyond the existing coverage.
A big thank you to Andrej Davidovič for building features with me and helping bring this release together.
Top comments (0)