Max.ai Analytics Assistant Release Notes - Version 25.01
Release Date: January 17th, 2024

Introduction

The Max.ai Analytics Assistant v25.01 introduces significant enhancements that streamline skill creation, improve testing workflows, and elevate user experience. A key highlight of this release is the introduction of Component-Based Skill Development, a modular approach that transforms how skills are built and managed. Additionally, a redesigned settings page and expanded test run options offer greater customization and control for users.

Key Improvements

New Component-Based Skill

We’re introducing a groundbreaking Component Skill model to redefine how skills are built and managed. This model focuses on creating skills entirely from modular components, replacing previous sequential skills. Here’s how it enhances skill creation:

• Modular Approach:
  • Add components from a predefined backend list.
    Bind nodes to each other to form logical flows.
    Map inputs with restrictions based on type.
    Assign input values directly (static or dynamic variables).
• Visual Flow Mapping: Skills are built in an intuitive, visual structure, making it easier to understand and debug.
• Enhanced Flexibility: This paradigm provides greater adaptability for building and customizing skills to fit specific use cases.
• Preview Functionality: A flow preview ensures you can verify your configurations before deployment.
  This update aims to improve the overall experience of skill creation, reduce development time, and streamline workflows for advanced users.

New Assistant Settings Page

The Assistant Settings page has been redesigned to enhance usability:

• Settings are now grouped into intuitive categories instead of being stacked together.
  This streamlined structure improves navigation and overall user experience.

New System Prompt Fields

We’ve expanded the system prompts functionality with new fields to provide greater customization:

• Added Fields: Capabilities, limitations, example_questions, and parameter_guidance have been integrated into skill properties.
• Persona Field in Copilot Settings: The new persona field is accessible via {{ pipeline.copilot.persona }} in system prompts.
  These enhancements offer users more control and clarity when managing skill configurations.

More Test Run Configuration Options

Users now have more focused testing options with the updated test run configuration model:

• Select Specific Questions: Run tests on single or multiple questions instead of all questions in a collection.
• Parameter Selection Focus: Option to skip skill runs, allowing users to focus testing on the skill and parameter selections with much faster test cycles.
• Test Naming: Easily Name tests while creating test runs for better organization.

More Warning Returned in API

For API users, context-dropped warnings are now supported directly via the API:

• Provides an indicator when context is being dropped in longer conversations, ensuring better awareness for users working programmatically with Max.

Tested Models

This release has been tested with the following models to ensure compatibility and stability:

OpenAI:

• Chat: gpt-4o-2024-08-06
• Narrative: gpt-4o-2024-08-06
• Evaluations: gpt-4o-2024-08-06
• Embedding: text-embedding-3-small

Gemini:

• Chat: gemini-1.5-pro-002
• Narrative: gemini-1.5-pro-002
• Evaluations: gemini-1.5-pro-002
• Embedding: models/embedding-001

Anthropic:

• Chat: claude-3-5-sonnet-20241022
• Narrative: claude-3-5-sonnet-20241022
• Evaluations: claude-3-5-sonnet-20241022
• Embedding: Not Applicable