UT Spark: FAQs

All currently affiliated Students, Faculty, and Staff have access to UT Spark.

If you fall into one of these groups and receive an error message saying that the administrator restricts certain access, please contact the Service Desk for assistance.

There are a couple of ways for submitting feedback, the best way would be to contact the help desk using the guide here: UT Spark: How to Get HelpPreview

Be sure to explain why the response was incorrect and include the exported chat.

The other way would be to use the feedback feature in UT Spark by clicking on the feedback icons.

Open

Remember that AI interactions may have inaccuracies, this should improve over time however providing feedback on Official Agents with a specific purpose will help the UT Spark team refine the responses.

Currently UT Spark uses some of the latest OpenAI models, however there are plans for expansion in the future!

Available Models as of August 18th 2025:

• gpt-4.1

• gpt-4o

At this time personal agents use gpt-4o.

Models release and update frequently, you can see which model you are using within the chat function by looking at the top left corner of the chat window.

Open

GPT-5 is not yet available in UT Spark. The UT Spark team will share an update when it becomes available.

Within the UT Spark platform, other users cannot see your chats unless you export them to a file and share them willingly.

On the administrative side, the UT Spark platform stores prompt information in an encrypted database within it’s backend. This database serves to provide continuity for users chatting with UT Spark, this is a configuration that all AI platforms use for functionality. UT Spark provides an additional layer of privacy because it does not allow this data to be used for training purposes regardless of the model used.

Interactions are logged for security and support, they are not publicly viewable or reused.

Please see: UT Spark: Usage & Data Privacy GuidePreview

Official Agents are agents created by UT Staff that are vetted and configured for a specific purpose. Agents may be created and shared by authorized departmental staff for a unit, or college, or created by the UT Spark team for general campus wide use.

Yes! Personal agents are available to everyone to create.

These agents are private to you, and are a lightweight version of Official Agents. They can be configured with knowledge sources and custom instructions.

Please see this guide: UT Spark: How to Create a Personal AgentPreview

For both personal agents and the “Chat with File” feature:

Tips:

• If you upload a PDF or Word document that is an image of a document (scanned etc.) - switch the document to the code interpreter on upload, this way it uses OCR.

Open

• If you are doing data analysis, best practice is to upload a .csv instead of an excel file or workbook.

Known issues with file uploads:

• Users may sometimes experience issues uploading files with extensions in uppercase, if this occurs rename the file extension to lowercase.

Example: image.JPG should be renamed to image.jpg.

• In Personal Agents, users may experience difficulty uploading multiple different file types at one time. If you experience this issue, try uploading them in batches or by file type.

• File Size Limit: 16mb

This is for both Personal Agent knowledge and Chat with File uploads.

Due to backend requirements, the current limits apply:

• Page Limit: 50 Pages

  • NOTE: This is cumulative and not per file. This means 50 Pages total across documents in a personal agent or “Chat with File” session.

• For Personal Agent knowledge the limit on uploads is currently 20 files.

Depending on your browser and security settings, UT Spark may need permission to use your microphone.

You may see a pop-up asking to allow microphone or recording access, approve it to enable speech-to-text.

If no pop-up appears, check your browser’s site settings to adjust microphone permissions manually.

Open

Example for Chrome

Large Language Models (LLMs) generate responses in real time by predicting likely words and phrases based on patterns they’ve learned, not by retrieving fixed answers. Because this process is probabilistic, the model may follow slightly different paths each time, even with the same question.

Small differences in context, such as what was said earlier in the conversation, can also influence the outcome. This variability is intentional, as it allows the model to be flexible and adapt its wording, examples, and level of detail to the situation.

.