New Features

GPT-5.1

• GPT-5.1 replaces GPT-5.0 across the platform.
• Model is in full General Availability (GA) and approved for PHI-restricted users.

Assistants

• Assistants can now be configured to use GPT-5.1 or GPT-4.1 as the selectable default model when using the assistant.

Email Notifications

• Weekly email notifications alert users that their assistants / chat or threads will be deleted in 30 days or less.

Default Model

• Ability for users to select which default they would like to use when interacting with general PhoenixAI chat.

Improvements

Retention Period Updates

• Additional emphasis included that the feature relates to permanent deletion, not archiving.

Accessibility Enhancements

Data Source Management Screen

• Corrected multiple contrast issues
• Replaced non-semantic elements with native elements
• Improved label clarity to align with accessibility standards

Other GUI Enhancements

• Removed the word "Beta" from all known occurrences
• Minor text clarifications and standardized terminology throughout the platform
• Updated browser tab title to include "University of Chicago"
• Applied consistent wording to sync job status messages

Fixes

• Addressed issue where files uploaded to an assistant occasionally failed to process
• Updated messaging to accurately reflect the state of the data source connection
• Standardized all sync-related messages to the same format and terminology

New Features

GPT-5 for assistants

• Users can now choose between GPT-4.1 and GPT-5 when interacting with an Assistant.
• GPT-4.1 provides faster responses and supports file uploads.
• GPT-5 offers deeper reasoning and more advanced task handling for complex workflows.

Deletion Notification Emails

• Users now receive weekly email notifications for assistants and threads that will be deleted within the next 30 days due to inactivity.
• Notifications reflect user-defined retention settings and system retention requirements.

Knowledge Source Integration

• Supports Box, Google Drive, and OneDrive utilizing only university-managed sources.
• Secure, one-way sync of selected folders only.

Data Sources Management

• Users can add, remove, and manage data sources in a dedicated screen.
• Each assistant can connect up to 5 data sources.
• A data source can be shared across multiple assistants.
• Data source contents sync automatically nightly or manually by user.
• Assistant creators will be alerted via email for sync failures or empty folders for data sources they created.
• Max 20 data sources per user (across assistants); max 5 data sources per assistant.
• Max 100 files per data source; files must be ≤ 1 million tokens or ≤ 20 MB.
• Only files that are that are accessible by the assistant creator are included.

GPT 5.0

• Added new GPT-5.0 model with prompt limit of 50 prompts per day aligned to GPT-4.
  • GPT-5.0 is restricted for PHI users.
• Updated the model selection dropdown UI to reflect these changes.
• Optimized for coding and complex reasoning.
• Offers significant improvements over both GPT-4.1 and the o3 models, primarily in reasoning, reliability, efficiency, and context handling

Export Assistants Functionality

• Allows users to export and download an assistant .JSON file backup of one assistant at a time.
• Exported data includes assistant logo/icon, instructions, description, conversation starters, settings, data sources, and metadata for attached knowledge files (not the files themselves).
• Export initiated from the Assistants screen; the user selects where to save the file.
• Exporting does not change the original assistant.

Import Assistants Functionality

• Users can import assistant .JSON files via the Assistants screen.
• Restores assistant details and configurations.
• Includes validation for data sources and all core setup fields.
• Excludes uploaded knowledge files, which must be manually added after import.

GUI & User Experience Enhancements

• Assistants screen displays data source name, type, ID, status, and last time of sync.
• Buttons to easily add or remove data sources from assistants.
• Folders may be searched for in addition to browsing for folders.
• Files not supported for data sources appear grayed out on screen.
• Data sources can be activated/deactivated without deletion.

Improvements

• System messages remain on screen until dismissed by user.
• Large file upload error message now offers recommended solutions.
• Token limit error message displays total token count and suggests ways to reduce token size such as converting PDFs to DOCX.
• Tooltips added for knowledge files to show full file names on hover.
• Conversation starters within assistants are now optional when creating or updating assistants.
• Consistent responses when uploading image files; images now generate a descriptive response like non-image files.
• Delete custom assistant message updated for improved clarity.
• The Retention Period selection screen has been redesigned to provide a more intuitive user flow, especially for PHI Restricted and Non-PHI Restricted users selecting retention settings more restrictive than system defaults.

Fixes

• Special characters in text files are now removed or replaced with blanks, ensuring successful file uploads and preventing false “File Empty” errors.
• Eliminated extra slashes in generated code so users can copy code directly from the response window and run it without further modification.
• Fixed issue causing knowledge files to disappear when removing and adding new files in assistants.
• High and major accessibility issues on the Data Source Management screen have been resolved to align with accessibility compliance standards.

GUI Fixes

• Updated Assistant Create / Update screens to clearly indicate required fields.
• Text across multiple screens has been revised for consistency and clarity.
• Additional minor GUI inconsistencies have been resolved.

New Features

Compliance Features

HIPAA Compliance for PHI: This release introduces HIPAA compliance support for Personal Health Information (PHI) through the creation of a new security role: PHI Restricted. PhoenixAI must meet the data security requirements defined by the University Medical Center for handling PHI. To help ensure compliance, the PHI Restricted role limits entitlements for users who access PHI.

Features not currently available to users who work with PHI:

• GPT model o3-mini
• Internet Search
• Share Thread
• Share Assistant With Everyone
• Share Assistant With Group

Splash Screen For PHI Restricted Users: A new login screen appears the first time a PHI Restricted User logs into PhoenixAI. The screen outlines usage conditions and requires the user to click Accept & Continue before proceeding. After initial acceptance, the splash screen will not appear on future logins.

• PhoenixAI Terms of Use for Health Information Users
  • By accessing this generative AI application, you acknowledge and agree to the following terms:
    1. LIMITATION: This system may not be used for clinical decision-making, patient care, or any purpose outside approved research protocols.
    2. APPROVED USE: This system is strictly for authorized academic researchers who have completed the required HIPAA and research ethics training.
    3. CONFIDENTIALITY: All interactions with this system containing health information must remain confidential. Do not share outputs, screenshots, or system access with unauthorized individuals.
    4. DATA SECURITY: While this system employs security measures to protect against health information leakage, you are responsible for ensuring appropriate data handling in your interactions.
    5. COMPLIANCE: Usage must comply with the University’s data security policies, and all applicable research protocols.
    6. MONITORING: System usage is monitored for system health, compliance and security purposes only.
    7. NO WARRANTY: This system is provided "as is" without a warranty of any kind regarding accuracy, completeness, or reliability of outputs.

Group Based Access for Assistants

• A ServiceNow request form is being introduced to request group-based access for Assistants.
• Note: Only faculty and staff who are not PHI Restricted Users are eligible to submit this form.
• Access requests involving large student groups (e.g., all students in a course) must be made by faculty or staff. Students may still create assistants and manually share access with a limited number of individuals (e.g., 10 classmates and their instructor).
  

System Initiated Deletion of Threads/Chats, Assistants, all Associated Files

PhoenixAI is committed to protecting user privacy, including the specific needs of protecting health information. To support this, all inactive threads, assistants, and related data for PHI Restricted users will be automatically deleted after 90 days of inactivity. These data retention requirements will begin mid September, 2025.

Threads/Chats and all Associated Files

• Threads/Chats and all Associated Files
  • Non PHI Restricted Users - 365 days from last response date
  • PHI Restricted Users - 90 days from last response date
• Assistants and all Associated Files
  • Assistant Creator is not a PHI Restricted User - 365 days from last activity date
  • Assistant Creator is a PHI Restricted User - 90 days from last activity date

Chat Retention

You can choose to automatically delete inactive threads and chats after a specified period of inactivity. When enabled, any thread/chat that remains inactive for the selected number of days will be permanently deleted and cannot be recovered. Use the Active toggle/slider to enable this feature and enter the number of days (1–90) for custom automatic deletion, click Update to save your settings. A chat is considered inactive if no new prompt has been entered during the specified time period.

Internet Search

PhoenixAI now includes an Internet Search feature that allows the assistant and individual chats/threads to access up-to-date information. When this option is disabled, responses are based only on existing knowledge and may not include recent events. When enabled, the assistant can search the web to provide more current answers, including clickable source links showing where the information came from. An assistant creator can turn off Internet searches from their Assistant to focus searches only on internal knowledge and uploaded knowledge files.

For example, if you ask about a recent news event:

• With Internet Search disabled, the assistant may not provide a relevant or updated answer.
• With Internet Search enabled, the assistant can retrieve and display information from recent sources

You can experiment with the Internet Search toggle to see how it affects the assistant's responses. Please note: Always validate the results and cross-reference the sources provided to ensure accuracy. The Internet Search setting follows your current configuration when creating a new assistant. By default, it is disabled. If you enable it, that setting will persist for all new chats and assistants until you choose to disable it again.

Please note that only the GPT-4.1 model has access to Internet Search. This feature is not available in the o3-mini model as seen in the images below.

Dark Mode

Dark Mode is now available. You can enable the Dark Mode theme by navigating to Settings > General and toggling it on. This feature provides a darker interface that may be easier on the eyes, especially in low-light environments. Dark mode theme can be set to be “System, Light, or Dark. This setting is connected to your PhoenixAI profile.

Improvements

This section details what's been enhanced to existing functionality of PhoenixAI in this release.

Azure Agents

• Changed AI agent to Azure Agent AI, a tool from Microsoft.

Update from GPT-4o to GPT-4.1

• The new standard GPT automatically selected when opening the PhoenixAI tool.
• Maintains the 50 prompt limit for users to ask frequent questions and gather information.
• Prompts are evaluated once and the resulting answers are provided to the user.
• Updated reasoning and evaluation of tasks
• Users may upload files to ask questions about.
• Users may create custom Assistants for specific localized sources of information.
• Expanded assistant sharing capability
  • You can now share assistants with specific individuals or with a group. For details on how to use this updated sharing feature, refer to the Share an Assistant section of the PhoenixAI Assistants: Create, Edit, Share, Delete an Assistant KB.

Recommendation: This model is most useful for anyone with a wide variety of questions that need to be answered quickly. It is also well-suited for working with structured, tabular data. Overall, this model performs well across most tasks.

Update from o1 to o3-mini

• Alternative model that must be selected via the dropdown menu.
• Maintains the 50 prompt limit for faculty, 10 prompt limit for all other users.
• Prompts will be evaluated multiple times by the model to ensure a correct response.
• This model does not support uploading files.
• Assistants cannot be created using the o3-mini model.
• Optimized for science, coding, and mathematical reasoning.
• o3-mini will not be available to PHI restricted users.
• o3-mini will not have access to internet search.

Recommendation: This model is most useful to professors and researchers who have limited questions that require precise answers. This model prioritizes response accuracy and advanced reasoning.

New Features

Review this section to see what new functionality has been added to PhoenixAI in this release.

Incognito Mode

• Incognito Mode Creation and Features: Users can create temporary threads in incognito mode, which are not stored or displayed in their history. A visible indicator will show that incognito mode is active, and users can access a description of the feature. An incognito session is limited to 30 minutes from when the first prompt is entered. However, a new session can always be created.
• Restricted Functionality in Incognito Mode: While in incognito mode, elements like the sidebar, Assistants, and custom buttons are disabled, but links such as PhoenixAI Help and the UChicago logo remain active. The sidebar will also be marked as disabled for screen readers. Lastly, incognito threads cannot be shared.
• Thread Deletion and Non-Restoration: Incognito threads and files will be deleted upon exiting incognito mode, closing the application, closing the browser tab, or at the end of 30 minutes. These deletions are permanent and cannot be restored or exported.
• Incognito Mode Limitations: Users are unable to switch from an existing thread to incognito mode or convert an incognito thread to a regular chat. Only one incognito thread can be active at a time, and new incognito threads can be created while the mode is active.

Assistants

• Formerly known as Custom MyGPTs, are now referred to as Assistants.

Thread Sharing

• When sharing a thread, the recipient will only be able to see the chat up to the point in time when the thread was shared.
  • All subsequent prompts that the original user enters will not be shared.
  • Only authenticated users can access shared threads, and they cannot edit the thread.
  • Additionally, the shared thread recipient will be unable to view any files created or uploaded.
  • The original author of the shared thread will delete all of the shared versions of that thread when deleting it.
  • Importantly, shared threads expire after 16 weeks from the shared thread link's creation date.

  Archive Threads

  • Users can archive threads to remove them from the sidebar, with the option to unarchive them from a settings menu.
  • Users can still delete threads they created from the sidebar menu.
  • Archived threads can be found and managed through the settings menu.
  • Archiving a thread will not affect the view of shared thread recipients, and the "Hide Thread" option will appear in the UI when archiving.

  Rename Threads

  • Users can rename threads in their thread history (sidebar menu) to assign a personalized title.
  • Renaming a thread created by the user will not affect the title of the shared version, as shared threads are not live.

  Hide/Unhide Assistants

  • Users can hide Assistants they created or those shared with them from the sidebar, with the ability to unhide them later.
  • Hidden Assistants can be found and managed by the user.
  • Hiding and unhiding functionality applies to both user-created and shared Assistants.
  • Hiding a shared Assistant from the sidebar does not affect the view of the user who received it.

  Settings

  • Settings Screen Overview: A new settings screen allows users to update preferences, including managing archived chats and customizing chat settings. It also includes sections for data retention and chat archiving.
  • Archived Chat Management: The settings screen allows users to view, search, delete, and unarchive archived threads. A confirmation message appears when deleting all threads.
  • Shared Thread Management: Users may view and delete threads shared with others or shared with them, with the option to delete threads shared by others.
  • Assistants Management: The settings menu includes options to hide or unhide Assistants the user has created or been shared with. Users can also delete their own Assistants or those shared with them. However, deleting an Assistant shared with a user only deletes their copy and not the original.

  Search Feature

  • Search Functionality: The search button allows users to find threads based on keywords, and searches only thread titles, not content. Search is case-sensitive.
  • Archived Thread Search: Search includes visible and archived threads.

  Chart/Graph Generation

  • Used to generate images of charts and graphs based on the data received, not to create images of people, places, or objects.
    • The data received must be in text or file format (i.e. .csv, Excel) and in a tabular format.
    • If asked to generate images other than charts and graphs (e.g. a picture of a park on a sunny day), users are recommended to use Copilot.

  Chart/Graph Output

  • Charts and graphs are generated and displayed onscreen with an option to download in PNG (.png) format.

  Image Analysis

  • PhoenixAI will analyze uploaded images, providing a description of the image. When prompted, PhoenixAI can also display ALT text in a separate, copyable output.
  • Supported image formats include PNG (.png) and JPG (.jpg).
  • GPT-4o is the only model capable of image analysis.
  • Only one image file (max 20 MB) can be uploaded at a time, and video files are not supported in this release.

New Features

Review this section to see what new functionality has been added to PhoenixAI in this release.

o1 initial release

• Alternative model that must be selected via the dropdown menu located at the top center of the page.
• Limitations for non-faculty users control the amount of information that can be acquired in one day.
  • 50 prompt limit for faculty, 10 prompt limit for all other users
• Prompts will be evaluated multiple times by the model to ensure a correct response.
• This model does not support uploading files (attachment icon does not display).
• MyGPTs cannot utilize the o1 model for prompts and responses. Custom MyGPTs are not created with a default model (e.g. o1).
• The o1 model is experimental in nature. Sensitive / confidential data, such as PHI and PII, should not be used when entering a prompt.

Recommendation: This model is most useful to professors and researchers who have limited questions that require precise answers. This model prioritizes response accuracy and advanced reasoning. To learn more about the new o1 model and how it differs from GPT-4o, read our PhoenixAI GPT Model Differences article.

Fixes

The following PhoenixAI issue has been addressed in this release.

• Implemented UI fixes to improve LaTeX mathematical equation rendering.
• Only an authenticated user can download a file that PhoenixAI has generated for them.
• The link to a file that PhoenixAI has generated will not be present moving forward as it will be a button now for the user to click on to the download the file.
• Previous files that were generated by PhoenixAI will not be downloadable to the authorized user. Users will need to recreate the file.

Fixes

The following PhoenixAI issue has been addressed in this release.

• Resolved a page rendering issue that caused some users to experience responses scrolling to the top of the screen, leaving the page blank.

New Features

Review this section to see what new functionality has been added to PhoenixAI in this release.

MyGPT initial release

• Ability to create personal MyGPTs to allow for a user to upload documents to provide their own specific knowledge base with their own data.
• Ability to share customize MyGPTs with anyone who has the link to allow for them to utilize someone else's MyGPT.
• Ability to access custom MyGPTs created by others (GPT must be enabled with a shareable link).
• Ability to distinguish between different interfaces (PhoenixAI, your own MyGPTs and shared MyGPTs).
• Visit PhoenixAI MyGPT: Create, Edit, Share, Delete a Custom GPT for step by step instructions and resources about MyGPT.

Improvements

This section details what's been enhanced to existing functionality of PhoenixAI in this release.

• Enhanced overall system stability and performance capabilities.
• Provided a listing of chat customizations recommendations to provide a guide to enhance user customization of responses.
• Enhanced user experience when generating response by adding citations.
• Implemented various accessibility improvements in PhoenixAI for a better user experience.
• Increased prompt context limit from previously 4 to now being 8 – this will refine the response generated on PhoenixAI interface.

Fixes

All of the following PhoenixAI issues have been addressed in this release.

• Improved system stability and performance capabilities.
• Fixed timeout issue for long running prompt.

Improvements

This section details what's been enhanced to existing functionality of PhoenixAI in this release.

• Enhancements to system stability and responsiveness.
• Updated the production app to run asynchronously for better concurrency.
• Updated the worker nodes to actively release memory after file uploads.

Fixes

All of the following PhoenixAI issues have been addressed in this release.

• Improved system stability and performance capabilities.
• Initially deployed the Azure Open AI model in US East due to an outage, but moved it back to US East 2 with other Azure services for better performance and security.