# April 7 ## Agents ### **Agent Component: New testing and evaluation capabilities** #### **LLM as a Judge for smarter evaluations** You can now use an LLM to evaluate your Agent outputs. Instead of relying only on deterministic checks, you can define custom evaluation prompts to assess quality, relevance, tone, and correctness. This allows you to validate scenarios that require judgment, such as whether a response answers a question or follows the expected tone. The LLM returns a clear **Passed** or **Not Passed** result for each test case, making evaluations more flexible and aligned with real-world expectations. Configure your LLM as a Judge now

#### **Annotations and Insights for better analysis and iteration** You can now add annotations directly to execution traces, including Tools, System Message, User Message, and Guardrails. This helps you capture why something worked or failed during testing. As you annotate multiple executions, Insights automatically identifies patterns across your findings. It highlights what’s working, what needs attention, top issues, and recommended fixes, reducing the need to review each execution individually. Annotate and generate your first insights

## Platform Improvements ### **Updates on the Consumption Model page: Full visibility of your contract and usage** We have updated the **Consumption** **Model** page to provide a more granular view of your usage and contract. Now divided into two intuitive tabs, **Contract overview** and **Detailed usage & costs**, the new interface allows you to track your credit lifecycle in real-time. You can now monitor and analyze costs down to the pipeline level. With the new breakdown tables, managing your budget and understanding the impact of GB-seconds and executions on your total commitment has never been easier. Learn more about the page updates

### **Visual comparison of pipeline versions (Visual Diff)** Turn hours of investigation into seconds of clear analysis. The new **Visual Diff** brings total transparency to your workflow: simply access the options menu on the deployed pipeline cards on the **Run page** to compare pipeline versions side-by-side. Instantly identify additions, removals, and configuration adjustments with a visual interface designed to speed up reviews and troubleshooting. Compare your pipeline versions

## Connectors ### **Marketo connector updates** * **Support for asset APIs** Manage your Marketo **Folders and Programs** natively with new GET and POST support. This update automates directory organization and simplifies the control of webinars and campaigns, allowing for agile, centralized management of your marketing assets at scale. * **Automated token refresh** Eliminate manual OAuth management and complex retry logic with the new **Automated Token Refresh** feature. Ensure continuous connectivity by automatically handling token renewals, providing a more resilient and maintenance-free integration for your Marketo pipelines. Upgrade your Marketo integrations today ### **New Double Braces function: FUZZYMATCH** The new `FUZZYMATCH` function simplifies string comparison by enabling approximate similarity searches within your Double Braces expressions. It identifies matches despite variations in spelling, accents, or casing, eliminating the need for complex data cleaning logic and making your integration pipelines much smarter and more resilient to typos. Learn how to use the new function ### **Capsule Update: Alphanumeric CNPJ** According to **Normative Instruction RFB No. 2,119/2022**, the Brazilian Federal Revenue Service has introduced the new **alphanumeric CNPJ** format, which may be issued starting in **July 2026**. This change requires software systems to be prepared in advance to support the new format, ensuring regulatory compliance and business continuity. To support our customers during this transition, **Digibee has updated the CPF/CNPJ validation capsule (cpf-cnpj-validator)**, which is now compatible with the new alphanumeric CNPJ format. **What do you need to do?** Simply **redeploy your pipelines that use this capsule**, and the new version will be automatically applied to the flow, with no additional adjustments required. This update ensures that your integrations are ready for the new CNPJ format in a simple, secure, and transparent way. If you have any questions or need support during the update, our [**technical support team**](https://app.gitbook.com/s/boT4qPJIk6PZotrxlJWL/help-and-faq/support) is available to assist. ## Digibee Academy ### **Introducing the course Agent Building: Foundations** We are introducing the course **Agent Building: Foundations**, now available on the **Digibee Academy**, with videos and audio in Portuguese. In this course, you will learn how to design and implement pipelines with multiple AI agents in a hands-on way, combining orchestration, external tool integration, and production best practices. Throughout the journey, you will: * Orchestrate multiple AI agents within a single pipeline * Connect agents to external tools via MCP * Apply production-ready guardrails * Optimize costs by selecting the right model for each task Start learning today ## **Bug fixes** * **Error when running a Dataset in the Agent Component:** We fixed a bug that caused an error during Dataset execution, preventing the execution details from being displayed. * **Executions did not persist after saving in the Agent Component:** We fixed a bug that caused executions to be lost in the Agent Component, even after being marked to be saved. * **Total execution values showing too many decimal places in the Agent Component:** We updated the Total column in the executions tab to display only two decimal places, improving readability. * **Error opening old execution groups in the Agent Component:** We fixed a bug that prevented listing or correctly viewing execution groups with legacy data. * **Issues saving and reopening tests and executions in the Agent Component:** We fixed a bug that caused inconsistencies when retrieving datasets, evaluations, and execution groups, as well as confusing evaluation logs.