Ender Turing Updates — 2026-07-02 to 2026-07-15
Important changes
Fine-tune LLM reasoning effort per feature
Category: fix
Importance: high
What changed: New controls in the LLM settings allow administrators to tune the reasoning effort or thinking budget for each AI feature independently.
Customer benefit: This provides granular control to balance AI response quality, speed, and cost, and also resolves issues that caused Ender GPT chat to fail on certain Azure deployments.
Highlights:
- Settings are now managed in the UI instead of via environment variables.
- Configure reasoning separately for features like Summary, AutoQA, and Ender GPT chat.
- The editor validates your settings upon saving to prevent runtime failures.
@-mention suggestions in comments restored
Category: fix
Importance: medium
What changed: Typing the "@" symbol in a conversation comment now correctly displays a list of users to mention.
Customer benefit: You can once again tag colleagues in comments to loop them into a specific conversation for review or follow-up.
Highlights:
- The user suggestion list appears reliably when commenting on a conversation.
- Tagging teammates works as expected in both the main conversation view and reused comment editors.
Safer error messages for scheduled chat runs
Category: fix
Importance: medium
What changed: Failed scheduled Ender Chat runs now show clear, user-friendly failure reasons instead of internal system details.
Customer benefit: This change enhances security by preventing the exposure of sensitive system information in error messages.
Highlights:
- Failure messages in scheduled chat run history are now curated and human-readable.
- Recognizable issues, such as quota limits, are translated into actionable messages.
LLM topics get an automatic description baseline version
Category: improvement
Importance: medium
What changed: Every LLM topic now includes an automatic baseline version generated from its original plain-text description.
Customer benefit: You can instantly revert to the original description-based topic definition or compare its performance against newer structured versions.
Highlights:
- This baseline version is created automatically and can be activated with one click from the version history.
- The baseline is evaluated alongside the active version, allowing for direct accuracy comparisons.
More reliable media file retention and cleanup
Category: fix
Importance: medium
What changed: The media cleanup process now correctly detects and reports failed file deletions instead of silently marking them as successful.
Customer benefit: This ensures your data retention policies are enforced reliably and that expired media files are actually removed from storage.
Highlights:
- Failed deletion jobs will now trigger retries instead of being ignored.
- Incorrectly configured retention settings are now handled more safely to prevent accidental data loss.
Other changes
More reliable call categorization
Category: improvement
Importance: medium
What changed: The underlying engine for call categorization has been updated to use a more structured format, reducing errors and improving consistency.
Customer benefit: You will see more reliable and predictable category assignments for your conversations.
Highlights:
- This reduces the number of failed categorization runs caused by malformed AI responses.
- A new "Single-topic Categorisation" managed prompt is now available for bucketing calls by their dominant purpose.
- The prompt editor now prevents saving configurations that would be guaranteed to fail.
LLM cost visibility for customer admins
Category: fix
Importance: medium
What changed: The "Enable LLM cost tracker" setting now correctly shows LLM costs to customer admins with the required permissions.
Customer benefit: Admins can now self-serve and monitor LLM-related spending directly within the platform when cost tracking is enabled.
Highlights:
- When enabled, costs are visible on the EnderGPT Chat page and the Prompts list in Analytics.
- Cost visibility requires both the global setting to be on and the user to have the "automation manage" permission.
Correct date filters on weekly chart drill-downs
Category: fix
Importance: medium
What changed: Clicking a data point for a specific week in a chart now applies the correct Monday-to-Sunday date filter to the conversations list.
Customer benefit: You can investigate weekly trends with a single click, landing on the correct set of conversations without needing to manually fix filters.
Highlights:
- This change eliminates the creation of empty or duplicate date range filters when drilling down from a weekly view.
- Weeks that cross a year boundary are now handled correctly.
Fast-track urgent recordings for processing
Category: feature
Importance: medium
What changed: You can now mark an uploaded recording for high-priority processing by including a special tag in its filename.
Customer benefit: This allows you to get analysis results faster for time-sensitive recordings by letting them jump ahead in the processing queue.
Highlights:
- This feature is opt-in and must be configured by an administrator.
- The high-priority tag overrides the automatic low-priority setting for very short recordings.
Dashboard remains stable when filtering by Email channel
Category: fix
Importance: medium
What changed: The main dashboard no longer becomes unresponsive when you filter the Session Duration widget to show only Email conversations.
Customer benefit: You can reliably filter your dashboard metrics by any channel without the page crashing or forcing a reload.
Initial category versions are now always scored for accuracy
Category: fix
Importance: medium
What changed: An accuracy score is now always calculated for a new category's first version, even if reviewers agree with all of its initial sample classifications.
Customer benefit: You get immediate visibility into a new category's performance, providing a clear baseline to measure future improvements against.
Highlights:
- This removes the confusing "no accuracy" state for new categories that perform well out of the box.
Campaign name filter added to automation triggers
Category: improvement
Importance: medium
What changed: The Campaign name is now available as a filter condition when creating or editing triggers for automations.
Customer benefit: You can now build more precise automations that run exclusively on conversations from one or more specific campaigns.
Highlights:
- This brings the trigger filters in line with the campaign filtering options available elsewhere in the platform.
Media retention policies now work for Azure Blob storage
Category: fix
Importance: medium
What changed: Audio files stored in Azure Blob storage are now correctly deleted according to your configured retention policy.
Customer benefit: This ensures that data retention rules are properly enforced on Azure deployments, helping control storage costs and meet compliance requirements.
Highlights:
- The cleanup process for Azure storage now behaves consistently with other storage backends.
Next run time for scheduled chats updates correctly after edits
Category: fix
Importance: medium
What changed: Editing the time, frequency, or timezone of a scheduled Ender Chat now immediately updates its displayed "next run" time.
Customer benefit: You can be confident that your scheduled reports will run at the correct new time after you make changes to their schedule.
Highlights:
- The guidance text for the "Defer next run" option has been clarified.
AI analysis is now skipped for empty calls
Category: fix
Importance: medium
What changed: AI-powered actions like Auto QA, Summary, and LLM Categorisation are now automatically skipped for short or silent calls that do not have a transcript.
Customer benefit: This prevents wasted processing and keeps your reports clean from meaningless AI results on non-conversational sessions.
Highlights:
- Automations based on metadata (like adding a comment or sending a webhook) will still run on these calls as expected.
- This change results in cleaner data for LLM categorisation training and evaluation.