Allow Users to Switch Models in Dash Enterprise

User Problem

Currently, Dash Enterprise administrators are limited to registering only a single AI provider and model within the AI Proxy configuration. This constraint forces a "one-size-fits-all" approach to generative AI tasks within Plotly Studio. Users who need to compare output quality across different Large Language Models (LLMs) or optimize for cost by using smaller models for simple tasks and high-performance models for complex reasoning cannot do so within a single environment. To bypass this, administrators would theoretically need to deploy multiple separate AI Proxy applications, creating significant overhead and a fragmented workflow for the end user.

What is it?

This feature is a configuration enhancement for the Dash Enterprise AI Proxy—the centralized application that facilitates secure communication between Plotly Studio and a customer’s chosen LLM provider. This update would transition the AI Proxy from a single-model gateway to a multi-model registry. It allows administrators to input and manage a list of multiple model endpoints and providers within a single deployed instance of the AI Proxy.

What does it allow users to do?

By enabling a multi-model configuration, the following capabilities are unlocked for users:

• Dynamic Model Switching: Users can toggle between different registered LLMs directly within the Plotly Studio interface to find the best fit for their specific data visualization or coding task.

• Performance Benchmarking: Users can run the same prompt against multiple models (e.g., switching between GPT-4 and Claude 3) to compare the accuracy, speed, and relevance of the generated code or insights.

• Cost Management: Users can select "lighter" or more cost-effective models for routine tasks and reserve more "expensive," high-parameter models for sophisticated data analysis or complex debugging.

• Centralized Administration: Administrators can manage all approved AI endpoints in one location, ensuring consistent security and governance across all models used by the organization without deploying redundant infrastructure.