Configuring LLM Settings & Tooling (VTA Setup Page 6)

Written By Asad Jobanputra

Last updated 3 months ago

Overview

This guide covers the LLM Settings page, where you control the advanced technical features of your VTA. This includes selecting the AI model, granting the VTA access to external tools (like websearch or code generation), and configuring debugging options.

Who it's For

Instructors: To enable or disable specific functionalities (like Image Generation).
LMS Admins/IT Staff: To select preferred LLM vendors (e.g., GPT 4o) and configure traceability for debugging.

Why Use It

These controls define the VTA's core intelligence and capabilities. They allow you to:

Optimize Performance and Cost: Choose the model that balances speed, accuracy, and institutional cost.
Expand Utility: Grant the VTA specific tools (like Code Interpreter) to make it useful for specialized courses (like programming).
Ensure Oversight: Enable Save Chat History for pedagogical review and auditing.

Part 1: LLM Available Models

This section allows you to select which AI models the VTA will use to process student requests.

Locate the LLM Available Models section.
Choose what models the LLM is available to use by selecting the corresponding toggle.
- Example: You can choose to run the VTA using either GPT 4o or GPT 4.1.

Note: The underlying vendor (e.g., azure_openai) is often displayed below the model name. Your available options are dependent on your institution's subscription and data governance policies.

Part 2: LLM Tooling

This section grants the VTA access to external functionalities beyond its core knowledge base. Toggle ON a tool to enable its capability.

Tooling Option	Action	Purpose and Impact
Websearch	Enable Websearch	Allows the VTA to gather current data from the web. Warning: Enabling this may introduce data external to your course materials, which can conflict with academic integrity guardrails.
Code Interpreter	Enable Code Interpreter	Allows the AI to generate code. Use Case: Essential for programming or data science courses where students need example code or syntax checks.
Image Generation	Enable Image Generation	Allows the AI to respond by generating images. Use Case: Beneficial for visual learners, art history, or science courses that need conceptual diagrams.

Part 3: Debugging Options

These controls are primarily used by administrators and instructors to troubleshoot issues and gather usage data.

Debugging Option	Action	Purpose and Benefit
Enable Traceability	Toggle ON	Allows saving of additional logs and function calls to troubleshoot agent retrieval. Use Case: If the VTA gives an inaccurate answer, IT can trace the exact file/step where the VTA pulled the wrong information.
Save Chat History	Toggle ON	This will enable professors to access student query information. Use Case: Essential for pedagogical review and using VTA Analytics to identify learning gaps across the course.

Best Practices / Notes

Limit External Tools: For most academic courses, it is recommended to disable Websearch and rely on the content provided in the Training Data to maintain the integrity of the course curriculum.
Performance vs. Cost: If you have multiple models available, you may choose a slightly older model (e.g., GPT 4.1) for simpler tasks to optimize resource usage, reserving the newest models (e.g., GPT 4o) for high-stakes courses.