April 29

Agents

Agent Component

LLM scorer variables and new templates

You can now reference dynamic values in the LLM scorer prompt using variables, allowing the judge model to evaluate the Agent's output in context rather than in isolation.

The following variable types are available:

• {{agent.userPrompt}} and {{agent.systemPrompt}} reference the prompts configured in the Agent.

• {{experiment.your-variable}} lets you declare custom variables in the scorer prompt and assign a specific value to each Test Case in your Dataset. The judge model then picks up those values at evaluation time, so each case is assessed with the right context.

New prompt templates are also available. Some already include predefined experiment. variables to help you get started faster.

Learn about the LLM scorer

Agent Component now available in Capsules

You can now use the Agent Component when building Capsules, bringing AI capabilities into reusable, modular pipeline components.

This allows you to encapsulate AI-powered logic and share it across multiple pipelines, making it easier to maintain and scale intelligent integrations throughout your organization.

Learn about Capsules

In-app guidance: Build your first AI agent

We've launched an interactive in-app tutorial to help you build your first AI agent directly on the Digibee Integration Platform — no prior experience required.

The tutorial walks you through creating an AI-powered pipeline that analyzes customer feedback sentiment using the Digibee Agent Component and returns a structured JSON response. By the end, you will have configured an Agent Component, set up conditional routing with the Choice connector, and tested a complete AI-powered pipeline end to end.

How to access the tutorial via Helphub:

1. Go to the Canvas area.

2. Select the ? icon in the left menu.

3. Select Tutorials e Checklists.

4. In the checklist that opens in the lower-right corner, select Build Your First AI Agent.

The tutorial is available to all users and takes approximately 10 minutes to complete.

---

Platform Improvements

Refreshed Capsules list screen

The Capsules list screen has been redesigned with an updated interface and a new list view layout, making it easier to navigate and manage your Capsules.

Go to Capsules

Updated Monitor screens

The Monitor screens have been updated with a new UI, delivering a more consistent and refined experience when tracking and managing your pipeline executions.

Go to Monitor

---

Documentation

We've created the following use case documentation to help you expand your knowledge of ZTNA Inverse Flow on the Digibee Integration Platform:

• How to add a Load Balancer to a ZTNA Inverse Flow: Learn how to configure a Load Balancer in your cloud infrastructure to distribute traffic across multiple Edge Routers, ensuring high availability and continuity for your ZTNA Inverse Flow integrations.

---

Bug Fixes

• Stream Parquet File Reader: Fixed a non-deterministic issue where date fields were returned as integers instead of formatted date strings during concurrent pipeline executions; date conversion now works reliably across parallel runs.

• Accounts in Build: Removed a misleading warning in the account creation modal that incorrectly advised against storing passwords and tokens in accounts, aligning the Build flow with official documentation and the Settings experience.

• Email V2 connector: Fixed an issue where existing pipelines lost their account references after a connector update; configured accounts are now preserved correctly without requiring reconfiguration.

• HTTP File trigger: Fixed a connection exhaustion issue where the HTTP File trigger was reaching its maximum connection limit, preventing users from connecting properly; connections are now handled efficiently without hitting the limit.