Release Date: 15-August-2024
AppExchange Link: Download GPTfy
New Features
- Prompt Builder: The Prompt Builder interface has been redesigned for improved usability. Users can now test prompts directly within the builder without publishing them first. This helps Prompt Engineers to iterate rapidly and test prompts without leaving the builder.
- Prevent Token Size spillover: GPTfy estimates the token size before invoking your AI model. It determines the maximum token size your model can process and avoids callouts if a mismatch is detected.
- Support for 2 Million+ Token Size AI Models: Newer AI models support very large token sizes, sometimes exceeding 2 million. To leverage this increasing token size, GPTfy now employs a file-based approach to handle large volumes of extracted records sent to the AI for processing.
- Support for Dynamic Locale: Users prefer to see dates and numbers formatted according to their locale. GPTfy now determines the user’s locale information and incorporates it into AI prompt grounding for improved readability. This support is also extended to AI Mass processing, where GPTfy processes records owned by different users.
- File Processing Guardrails: Users can append files to an AI prompt for processing. GPTfy now includes guardrails to limit the use of files. Admins can configure the maximum file count, size, and type allowed in file processing prompts.
- Support for AWS LlaMA: GPTfy now supports Meta’s LlaMA large language model hosted on AWS or other standards-based platforms.
- Process Salesforce History Records: Derive deeper insights by including records from the Salesforce history object. This release introduces the ability to add the history object as a child object within the Data Context mapping. This functionality supports any Standard, Custom, or Managed Package object where history tracking is enabled at both the object and field level.
- Preferences Tab: Admins can now manage GPTfy preferences directly from the Preferences tab. This eliminates the need to modify Custom Settings for GPTfy configurations.
- Custom Hourly Cost Definition: GPTfy ROI reports now allow you to input your organization’s average hourly cost. This enhancement provides more accurate calculations of monetary savings realized from time saved through AI utilization.
Enhancements
- Snowflake Integration: GPTfy has been optimized to integrate with Snowflake through the Salesforce Lightning connector (OData-based connector).
- Data Retention Optimization:
- The retention configuration now supports granular purging of specific security audit records. This allows for the retention of critical exception records while efficiently managing data storage and security posture.
- Scheduling retention jobs has been simplified with the introduction of a new scheduler modal window. This window appears when retention is enabled, eliminating the need to create Apex jobs using Cron expressions.
- Query Builder: GPTfy leverages a shared Query Builder component to define custom filter criteria (SOQL WHERE conditions) for several features. This builder also allows users to specify optional order by and limit parameters. All filter criteria are instantly validated to ensure they adhere to SOQL syntax. The enhanced Query Builder is now accessible within the following GPTfy features:
- Data Retention
- Data Context Mapping
- AI Mass Processing
- Prompt Visibility
- AI Callout Timeout Enhancements: The callout architecture has been modified to enable the creation of Security Audit records in the event of timeout issues. Timeout exceptions will now be captured within the Security Audit record, with detailed information logged on the associated exception records.
- GPTfy User Console Optimization:
- Handling Large Responses: GPTfy User Console now handles large responses more effectively. A vertical scrollbar is automatically added when responses exceed the console’s standard height, allowing users to easily navigate and view the full content.
- Disqualified Prompts: GPTfy button on the GPTfy User Console will be disabled (greyed out) for records where the selected prompt is ineligible. A tooltip will display the specific reason why the prompt is not available for the current record. This enhancement provides transparency to end users about prompt disqualification.
- Security Audit Optimization
- Enhanced Response Display: The Security Audit Response tab now presents fully formatted responses. For HTML responses, users can view both the rendered HTML and the raw HTML to facilitate efficient troubleshooting.
- Improved Page Layout: Page layout modifications have been made to relocate show/hide fields for better visibility and organization.
- Ability to pull Account Activities: Account activities have historically been linked differently in Salesforce than activities for other objects, preventing GPTfy from extracting them. With this release, GPTfy can now extract Account activities by leveraging the data context mapping configuration.
- Ability to add FeedItem as a child: Data Context Mapping now supports adding a FeedItem (Chatter) as a child to any standard, custom, or managed package object where FeedItem is enabled.
- Availability of API Data Source for AI Mass Processing: AI Mass Processing now supports custom pre-processing and post-processing logic through API Data Sources. This enables flexible handling of AI callouts, even when processed in bulk, by allowing users to implement custom code for data manipulation.
- Improved Fault Tolerance for AI Response Data Types: AI responses often exhibit inconsistencies in data type formatting, such as numbers being returned with or without quotation marks. This variability can hinder GPTfy’s ability to accurately populate Salesforce fields. With this release, GPTfy enhances its fault tolerance by optimizing AI-returned values to ensure compatibility with corresponding Salesforce field data types.
- Security Layer Optimization:
- Regex-based Security Filters: GPTfy now utilizes optimized algorithms to automatically determine the optimal order for applying Regex-based Security Filters. This enhancement improves processing efficiency and prevents conflicts between multiple filters.
- Blocklist-based Security Filters: Blocklist-based Security Filters are now case-insensitive, ensuring that any value added to the blocklist is masked before being sent to the AI for processing.
- Unsupported Fields on Data Context Mapping: Some fields on feed items are inaccessible through SOQL. To prevent exceptions, Data Context mapping now hides these inaccessible fields from the user interface. Previously, selecting these fields could have caused errors.
- Prevent Duplicate AI Model: Implemented validation to prevent duplicate AI Models from being created during prompt imports using the GPTfy Import wizard.
- File Attachment Warning for GPTfy Gmail Plugin: The GPTfy user console within the Salesforce plugin for Gmail and Outlook now displays a warning when a prompt configured with ‘Include Files’ set to true is missing a file attachment.
- Timestamp on Target field for JSON Prompts: A timestamp can now be optionally added to the Target field, even when using a JSON Prompt. If “Add Timestamp” is selected, GPTfy will prepend the timestamp to the JSON response.
- Prompt Scheduler Window Alignment: AI Mass Processing’s prompt scheduler windows (create and edit) now utilize the same Custom Prompt scheduler window for consistency.
Bug Fixes
- Usage Tracking Dates: Fixed issues with Usage Tracking Detail records not reflecting the correct start and end dates.
- Prompt Reinstatement: Previously, reinstating an older version of the Prompt reverted the prompt name to its original state when the version was created. With this release, reinstating Prompt restores the prompt and all associated information, but the prompt name remains unchanged.
- ROI Dashboard Metric: The ROI Card on the GPTfy Cockpit displayed an incorrect Amount. This issue has been resolved.
- Post-Install Script Fix: Previously, GPTfy installations resulted in Data Context Mapping records being created in a disabled state, causing subsequent issues. This problem has been resolved. New GPTfy installations now generate Data Context Mapping records with an active status.
- Query Builder Number Field Handling: Previously, number literals entered into the Query Builder were treated as text values. This release enhances the Query Builder’s data type awareness to correctly handle numbers as numeric values.
- Prompt Catalog Purpose Cards: The Catalog Folder name update was causing issues. Prompt names are now read-only to prevent syncing issues.
- REST API Prompt Error: Enhanced error handling for REST API calls with User Input enabled. The system now explicitly returns the reason for failure when no input or prompt command is provided.
- AI Model Record Deletion: Previously, deleting an AI Model card did not also delete related dependent information. This issue has been resolved. Now, when an AI model is deleted, GPTfy will delete all associated records to maintain data integrity.
- Prompt Activation Issue: Resolved an issue where Prompt activation failed when the Prompt component was placed on a child object and an AI processing action updated a field on the parent object.
Known Issues
- Access Issue on Gmail Plugin: When logged in as a Standard user, the Sample Visualforce component included in the package has an access problem. However, it functions correctly when a new Visualforce component that is identical to the sample one is created.
We’re always working to make things even better!
The problem that GPTfy is solving is to help you leverage the power of Generative AI in your Salesforce environment without any complex coding setup or worrying about security!
With GPTfy, you can create and execute AI prompts that can generate high-quality and relevant content for your business needs quickly and easily.