Release notes: Oct 16, 2025 (v9.42.1)

Introducing our new API reference documentation

We're excited to announce a complete overhaul of our API documentation, designed to provide a better experience for developers. We've focused on adding more code examples, better explanations, and a more intuitive structure.

Here's a look at what's new:

• Public access: The API documentation is now open and accessible to everyone without requiring a login.

• New "Get Started" guide: Jumpstart your integration with a dedicated guide that includes complete, copy-paste-ready code examples.

• Core concepts explained: We've added detailed explanations for core concepts like Roles, Scanners, Campaigns, and Reports, complete with code samples to show you how they work.

• Redesigned UI: The new interface makes it easier to navigate our API. We've improved the explanations for every endpoint, so you know exactly what to expect.

• Product-specific sections: Find what you need faster with new sections dedicated to our Inference Defend and Inference Red-Team products.

• Chatbot integration guides: Learn how to connect with our platform through our Slack and Microsoft Teams chatbots.

• Troubleshooting section: Quickly resolve common problems with our new guide to frequent issues and their solutions.

• Link to product docs: Easy access to our main product documentation for more in-depth information.

Check out the new API documentation here: docs.calypsoai.com.

AI Guardrails

Test scanners with a custom dataset

This release introduces the ability to evaluate scanners using your own dataset of prompts, giving teams a reliable way to measure accuracy, detect false positives and negatives, and fine-tune scanner performance before publishing.

Users can now:

• Upload a custom CSV dataset containing prompts and their expected outcomes (true for prompts that should be blocked, false for prompts that should be allowed).

• Automatically validate uploaded datasets – checking file size, header names, column count, and label format.

• Run tests against specific scanners and versions, or compare multiple versions at once.

• View detailed performance metrics, including:

  • F1 score (when both true and false labels are present)

  • Confusion matrix with true positives, true negatives, false positives, and false negatives

  • Overall accuracy

• Download a detailed CSV report showing scanner decisions for each prompt (available for 7 days).

• Track test history in the Dataset tab, including test name, dataset used, scanner versions tested, and timestamps.

To run a test, go to the Playground, click Upload dataset in the top-right corner, and follow the guidance in the upload wizard. After the test completes, click View dataset results in the success message to open the report.

Each test appears as a card in the Dataset Results tab, showing progress or the final F1 score. From there, click Details to view the full confusion matrix and performance breakdown.

The most recent F1 score also appears in the Version history panel of that scanner, making it easy to see how performance changes across versions.

View projects in table or card layout

This release introduces a new table view for Projects, giving users more flexibility in how they view and manage their workspace.

You can now toggle between the card view and the new table view using the layout switcher icons in the top-right corner. Both views share the same filters and search funcationality; filters applied in one view automatically carry over to the other.

This update improves visibility for large environments and makes it easier to compare project metadata side by side.

Redacted content now returned in Scans API responses

The /scans API now includes redacted content directly in the response when a scanner with a redact action is triggered. This enhancement:

• Eliminates extra API calls: The redacted prompt is now in the original response, removing the need for an additional request to fetch the sanitized data, especially in out-of-band deployments.

• Reduces API load and simplifies integration.

• Ensures consistency for subsequent processing steps.

AI Red Team

Quick view of mitigation actions

This release introduces a new column in the Reports table to give you a faster, at-a-glance summary of the volume of recommended actions for each campaign run.