September 2025

Markdown Editor in Note Nodes

September 23rd, 2025

Note Nodes can be an effective way to document your Workflow and communicate different aspects of it. Historically, you could change the font size of the note’s text, as well as its background color, but that was about it.

Now, Note Nodes support markdown. You can add headers, style text, include code blocks, and even embed iframes. Coupled with the recently added ability to move Note Nodes behind other nodes, you can really get creative with how you document your Workflows for others to follow along.

AI App UI for Deployed Workflows

September 22nd, 2025

Deployed Workflows now feature a dedicated AI App UI that enables users to execute and test Workflows directly within Vellum’s web interface. This chat-based interface provides a first-class testing experience without requiring API integration, making Workflow testing more accessible and streamlined than ever before.

Why It Matters

Previously, testing deployed Workflows required developers to implement API calls in their applications or use external tools like Postman to execute Workflows. This created barriers for non-technical team members who needed to test Workflow behavior, validate outputs, or demonstrate functionality to stakeholders. The new AI App UI eliminates these friction points by providing a dedicated, intuitive interface for Workflow execution within the Vellum platform itself.

This development Workflow improvement means faster iteration cycles, broader team access to Workflow testing, and simplified demonstration capabilities that don’t require technical API knowledge.

Key Capabilities

The AI App UI provides a comprehensive testing environment that mirrors your production Workflow execution while remaining user-friendly and accessible:

• Intuitive Input Interface: Present clean, automatically-generated input forms that match your Workflow’s expected parameters, including support for text inputs, numerical values, file uploads, and other data types your Workflow requires
• Real-time Execution: Execute Workflows with the same reliability and performance as API calls, with streaming output support that shows results as they’re generated
• Interactive Testing Flow: Edit inputs and re-run executions multiple times within the same session, enabling rapid testing and refinement without leaving the interface
• Direct Access from Multiple Entry Points: Launch the AI App UI directly from the Share Workflow modal or through the Deployments overview page, integrating seamlessly into existing Workflow management workflows

How to Access

The AI App UI can be accessed through two convenient entry points:

1. Share Workflow Button: Click the share button on any deployed Workflow to reveal sharing options, including the new AI App interface
2. Deployments Overview: Navigate to your Deployments overview page where deployed Workflows now display AI App access options alongside existing management tools

How It Works

The AI App UI automatically analyzes your deployed Workflow’s input requirements and generates an appropriate testing interface. When you execute a Workflow, the system calls the same Vellum APIs that power production integrations, ensuring test results accurately reflect real-world behavior. Streaming outputs appear in real-time, and you can modify inputs and re-execute as needed for thorough testing.

Interactive Components in Agent Builder

September 22nd, 2025

Agent Builder now supports rendering rich, interactive UI components directly within chat responses. This feature enables Agent Builder to present more sophisticated interfaces beyond plain text, allowing users to perform actions like creating environment variables, accepting Workflow plans, and running Workflows all from within an Agent Builder conversation.

Why It Matters

Previously, Agent Builder conversations were limited to plain text exchanges, requiring users to navigate away from their chat session to perform related actions or make configuration changes. This Workflow interruption broke the conversational flow and reduced productivity. With interactive components, Agent Builder becomes a comprehensive command center that keeps users engaged and productive within a single interface.

Key Capabilities

This feature enables Agent Builder to present sophisticated user interfaces that go far beyond traditional chatbot responses. Users can now:

• Create and manage environment variables directly from chat responses, streamlining development Workflows
• Review and accept Workflow plans with interactive approval controls, enabling faster iteration cycles
• Execute Workflows with one-click actions embedded in the conversation, reducing context switching
• Perform complex actions through rich UI components rather than navigating to separate pages

How It Works

Interactive components are rendered as part of chat responses and include all the controls and inputs necessary for users to complete their tasks. When a user interacts with a component, the corresponding action is performed in real-time and Agent Builder receives feedback about the interaction to continue guiding the conversation.

Dynamic Model Recommendations on Agent Builder

September 17th, 2025

Agent Builder can now intelligently recommend models for your workflows, selecting the best options based on their capabilities, context windows, performance, and cost based on your specific requirements and use case.

Custom Node Colors & Icons

September 17th, 2025

You can now customize the color and icon for any Node in your Workflows! Simply click on a Node’s icon to open the customization interface, where you can choose from a variety of colors and search through a comprehensive icon library to find the perfect visual representation for your Node’s functionality.

SDK Preview Code Search

September 17th, 2025

You can now use the search bar in the SDK Preview to match on content! Search results appear with highlighted search terms and enable quick navigation to specific files within your SDK codebase. The search functionality finds matches in both file names and file content, making it easier to locate and navigate to the code you need.

Custom inputs for Agent Node code tool

September 15th, 2025

You can now pass inputs to the code tool that will not be populated to the model.

In the example below, input_1 is passed as an input, and the model will only populate greet. During tool calling, both inputs will be passed into the function.

Raw Data for Provider Errors

September 15th, 2025

You can now see the raw response for Provider Errors that might show up when executing a Prompt. This is particularly useful when you want more context as to why a particular Prompt request to a model provider failed.

Open Definition Button for Custom Nodes

September 11th, 2025

You can now open the Code Preview for your given Custom Node with this new button. Previously, you had to manually click the Code Preview button and navigate to the specific file for the Custom Node you are interested in. Now, you can save time clicking and searching with this button.

Agent Builder (Beta)

September 16th, 2025

Agent Builder is a conversational AI assistant that helps you build and optimize agent Workflows directly within Workflow Sandboxes. This new Beta feature transforms workflow creation from a manual process into an interactive conversation.

When you create a new Workflow Sandbox, Agent Builder appears as a dedicated panel ready to help you describe what you want to build. Simply tell it your requirements—like “create a workflow that extracts key points from a document and finds relevant quotes”—and watch as it constructs your workflow step by step.

As Agent Builder works, it shows you real-time progress updates, indicating which Nodes it’s configuring (Map Nodes, Search Nodes, Custom Nodes, Prompt Nodes, and more). You can watch as it completes each step of the workflow construction process.

The workflow updates automatically appear in your Sandbox, complete with pre-filled sample data and a Run button for immediate testing. You can easily see the created workflow alongside the Agent Builder panel and start experimenting right away.

We’ve been using Agent Builder internally to create powerful automations for transcribing calls and documents, converting them into Linear tickets, and even reviewing GitHub PRs. We’re excited to see what the community will build with this conversational workflow creation experience.

First-Class Image and Document Input Variables

September 10th, 2025

Previously if you wanted to pass an image or document into an LLM in Vellum, say to use the Vision capabilities similar to GPT-4 or GPT-5, you had to pass it as a Chat History variable type. This is unintuitive and difficult to set up and use. Often, you don’t want to pass a Chat History, just a file.

Now, both Prompts and Workflows have first-class support for Image and Document variable types in their Input Variables.

Static Images & Documents in Prompts

September 10th, 2025

You can now attach static images and documents directly to your Prompts without needing to include them as part of a chat history input variable. This makes it much simpler to provide consistent reference materials that should be sent with every Prompt execution.

To attach a static image or document, click the paperclip icon in the Prompt editor and choose “Insert Static Attachment”. The attached files will be included automatically with every execution of your Prompt.

This feature is particularly useful when you have standard reference documents, product images, or other materials that need to be consistently included across all executions of a particular Prompt.

Stack Trace for Workflow Execution Errors

September 8th, 2025

You can now see stack traces for any errors that show up when you execute your Deployed Workflow. This is particularly useful when you want more context as to why a particular Execution failed by giving you the error message and the Stack Trace of the error that was surfaced.

Create Custom Nodes from the UI

September 8th, 2025

You can now create Custom Nodes by drag-and-dropping them from the Node side panel.

Custom Nodes allow you to define any custom behavior by specifying:

• Input attributes
• Outputs to return
• And the underlying code to execute

Previously, you could only create new Custom Nodes by pushing up the code representation of a Workflow that contained one via the vellum workflows push CLI command. Now, you can define the Custom Node and edit its underlying code straight from the UI. To do this, click the Code Preview button, toggle on edit mode, and edit the definition of for your Custom Node.

Positioning Nodes In Front / Behind One Another

September 7th, 2025

You can now right-click on any Node in your Workflows to position it in front of or behind other Nodes. This feature is particularly useful when working with Note Nodes, allowing you to place them behind other Nodes to help document and organize different parts of your Workflow.

This layering system makes it easier to create visually organized Workflows where documentation and workflow logic can be cleanly separated, improving both readability and collaboration.

Updated Data Retention Policy for New Organizations

September 4th, 2025

New Organizations now default to a 30-day retention period for monitoring data, helping optimize storage costs and performance while maintaining essential operational insights. This change provides a balanced approach to data management without impacting functionality.

Existing Organizations maintain their current data retention settings, and Organizations on paid plans can still customize their retention periods to meet specific business requirements. If you need longer data retention for your use case, please reach out to our support team who can help configure the appropriate settings for your needs.

Syntax Highlighting in Prompt Jinja Blocks

September 1st, 2025

Prompt Jinja Blocks now include line numbers and syntax highlighting, making it significantly easier to write and validate Jinja code.

Previously, Jinja blocks appeared as plain text without visual feedback, making it difficult to spot syntax errors or understand code structure.