Overview
Gen AI Studio is where you create and manage the AI prompts that drive Ender Turing's generative AI features. Each prompt is a set of instructions sent to a large language model (LLM) to perform a specific task on your conversations — such as generating summaries, classifying topics, scoring quality, or powering EnderChat analytics.
You can maintain multiple prompts per category, inherit the category's model or select a provider and model, and track prompt versions over time.
How It Works
Accessing Gen AI Studio
Expand Analytics configuration in the main navigation and select Gen AI Studio.
Who Can Use Gen AI Studio
Action | Requirement |
View, create, edit, archive, duplicate, or delete prompts | Automation manage permission |
Users without the required permission cannot access Gen AI Studio.
The Auto-QA and EnderChat tabs are visible only to users with an extended platform permission. Other users see only the Summarisation and Categorisation tabs.
Prompt Categories
Prompts are organized into four categories, shown as tabs at the top of the list:
Category | Purpose |
Summarisation | Generate text or structured JSON summaries of conversations |
Categorisation | Automatically classify conversations into topics or categories |
Auto-QA | Evaluate conversation quality against scorecard criteria |
EnderChat | Power the EnderGPT chat interface for conversational analytics |
Your selected tab is remembered between visits.
Ender Turing Managed Prompts
Some prompts are maintained directly by Ender Turing and ship with the platform. They appear in the prompt list with Ender Turing as the author and an ET Managed chip on the row.
Managed prompts cover ready-to-use scenarios such as:
Default Auto QA Prompt — runs in the Auto-QA category. It scores each selected scorecard point as satisfied, not satisfied, or not applicable, captures supporting agent quotes, and explains the decision in the conversation language. The output is structured (JSON) so that scores feed into scorecards automatically.
Default Categorisation (multi-topic) — runs in the Categorisation category. Evaluates every category selected by an Ender categorisation action against the conversation and assigns all categories that are clearly justified by the transcript. Best for calls that can legitimately span more than one topic. Every category that matches is reported, so no assignment is lost even when the Ender selects a large number of categories; alongside the matches it also reports the 10 non-matching categories that came closest to matching, and leaves the remaining non-matches out of its answer. Categories it leaves out were still evaluated — they simply did not match.
Single-topic Categorisation — also in the Categorisation category. Evaluates every category the same way but assigns at most one — the primary and dominant purpose of the call. Secondary or incidental topics are ignored. Use it when each conversation must land in exactly one topic for reporting.
CSAT Dynamics and Interaction Intensity Dynamics — Summarisation prompts that power the matching dashboards.
What you can and cannot do with a managed prompt:
Action | Available on managed prompts |
Open the prompt to view its name and description | Yes (eye icon on the row) |
See which Enders use it (usage panel) | Yes |
Edit the text, model, or options | No |
Archive, restore, duplicate, or delete | No |
See version history | No — managed prompts are not versioned in Gen AI Studio |
When you open a managed prompt, the editor shows a short read-only summary (name, description, and an "Managed by Ender Turing" note) instead of the full prompt editor. To request changes or improvements, contact Ender Turing support.
The Auto-QA category is managed by default: the Default Auto QA Prompt is what Enders' Auto QA (Gen AI Based) action runs unless an authorized user has selected a custom Auto-QA prompt. This means Auto-QA works out of the box without requiring you to author a prompt.
Categorisation works the same way. Both Default Categorisation (multi-topic) and Single-topic Categorisation are ready to select from an Ender's categorisation action — no prompt authoring required. Custom Categorisation prompts remain available for anyone who needs to tailor the wording, and appear alongside the managed ones in the same list.
Viewing the Prompt List
The list shows all prompts in the selected category. Each row displays:
Name — the prompt name
Version — the current version indicator
Usage — how many Enders (automations) reference this prompt
Updated — when the prompt was last modified
Last edited by — who made the last change
Cost 24hrs / Cost 30d — the prompt's USD spend over the last 24 hours and 30 days, visible to users with Automation-management permission. Zero spend is shown as
$0.00; unavailable costs are shown as--.
Use the Active / Archived toggle in the top bar to switch between active and archived prompts. Use the search field to filter prompts by name.
Creating a Prompt
Select the prompt category tab.
Click Create in the top bar.
If user-created prompts already exist in that category, the system duplicates the most recent one as a starting point. If only managed prompts exist or no prompts exist at all, a blank form opens (managed prompts are never duplicated).
Enter a Name (required) and optional Description.
Write the prompt text — the instructions the LLM will follow.
Use Default, or select a provider and model if your permissions allow it (see Model Configuration below).
Optionally enable Force JSON mode to constrain the LLM output to valid JSON.
The Use this Prompt Version checkbox marks this version as the release (production) version.
Click Create to save.
Editing a Prompt
Click the edit (pencil) icon on any active prompt row.
Modify the name, description, prompt text, model, or options.
Click Update to save.
Saving with content changes creates a new version — previous versions are preserved in the version history panel on the right side of the form. If the only change is toggling the Use this Prompt Version checkbox, the existing version is promoted to release without creating a new version.
Standard and Expert Editing Modes
Click the gear icon in the top-right corner of the prompt form to switch between:
Standard mode — a rich text editor with clickable variable buttons for inserting template variables.
Expert mode — a raw text editor with Jinja2 syntax highlighting, giving full control over the prompt template. A link to the Jinja2 template reference is provided.
Both modes edit the same prompt, but they may show different amounts of it, depending on whether the prompt text contains <standard-mode> markers.
Prompt with
<standard-mode>markers — typical for Ender Turing's built-in templates and for prompts where you added the markers yourself in expert mode. Expert mode shows the whole template, while standard mode shows only the instruction section between the markers. Anything outside the markers stays untouched while you edit in standard mode, so switching back and forth does not discard expert-only content.Prompt without those markers — typical for a prompt you created from scratch in standard mode. Both modes show the same text.
Standard mode is a rich text editor rather than a raw text editor, so XML- or HTML-style tags are not preserved in the part of the prompt it edits: switching to standard mode can strip them and save the reduced text. Keep such tags outside the <standard-mode> markers, or edit that prompt in expert mode only.
Prompts that do not include the transcript
Summarisation, Categorisation, and Auto-QA prompts need the conversation transcript. If a prompt does not include it — which is the normal result of writing a prompt from scratch in standard mode — Ender Turing adds a surrounding template when the prompt runs, taken from another released prompt in the same category, or from the category's default template if there is none.
Treat this as a safety net rather than something to rely on:
The instructions the prompt actually runs with are not the ones you see in the editor, and they can change when someone releases another prompt in the same category.
After the prompt has run, the version you open in the editor may already contain the added template.
EnderChat prompts work from prepared summaries instead of raw transcripts, so they are never extended this way.
To stay in control of the full instructions, write the complete template in expert mode — including the part that renders the transcript — and put <standard-mode> markers around the section you want to edit in standard mode. Adding the markers on their own is not enough; it is the transcript that makes the template complete.
Line breaks and pasted text
Standard mode keeps the structure of what you type or paste:
Press Enter to start a new line inside the prompt.
Paste multi-paragraph text — for example from Microsoft Word or another document editor — and the paragraph breaks are kept when you save and when you reopen the prompt later.
Blank lines at the very start and end of the prompt are removed when you save. Blank lines in the middle are kept.
If you open an older prompt version and see visible \n characters in the text where line breaks should be, that text was stored that way by an earlier version of the editor. Existing versions are never rewritten automatically, because some prompts legitimately contain that character sequence. Remove the stray \n characters, press Enter where you want the breaks, and save — the new version stores real line breaks.
Template Variables
In Standard mode, click a variable button to insert it into your prompt. Available variables:
Variable | Description |
| The conversation type (call, chat, email, etc.) |
| The organization's configured language, used to tell the LLM which language to reply in |
In Expert mode, type variables directly using Jinja2 syntax (e.g., {{ session.type }}).
{{ language }} always renders the language configured for your organization, not the language detected on the individual conversation. This keeps analytics output (CSAT, intensity indicators, AutoQA, custom summaries) consistent across calls, even on bilingual conversations where per-call language detection can vary.
Model Configuration
If your organization allows model selection, choose a provider and model for the prompt, or select Default to follow the category's model configuration. An explicit selection applies to that prompt; Default follows future changes to the category configuration.
For Azure, provide the deployment name, endpoint, credentials, and optional pricing identity.
When model selection is restricted by system settings, only users with the extended platform permission can change it. Other users inherit the model from the previous version or the category default.
Category-Specific Parameters
Some categories have additional configuration options that appear below the main form fields. These parameters are specific to how the prompt is used within that category's workflow.
For Summarisation prompts, a Summary results visible to setting appears below the prompt editor. You can choose:
No restrictions (visible to all roles) — any user who can view the conversation can see summaries generated by this prompt.
Select specific roles — only users whose role is in the selected list can see the summaries. Use the role picker to choose one or more roles, or click Select All to include every role.
Version History
When editing an existing prompt, a version history panel appears on the right side. It shows all saved versions with:
The version number and creation date
Who created each version
Which version is marked as the release version (shown as Active)
Select any version to load it into the form. From the version history panel you can:
Promote a version — hover over a non-release version and click the checkmark icon to mark it as the release version. You can also check Use this Prompt Version in the form and save.
Delete a version — hover over a non-release version and click the delete icon to remove it permanently. The release version cannot be deleted.
Prompt Usage
Below the version history, a usage panel shows which Enders (automations) currently reference this prompt. This helps you understand the impact of changes before saving.
Duplicating a Prompt
Click the duplicate icon on any active prompt row. After confirming, a copy is created with the name appended with "- copy" and you are taken to the edit form for the new prompt.
Archiving and Restoring Prompts
To archive an active prompt, click the archive icon on its row and confirm the standard archive prompt.
To restore an archived prompt, switch to the Archived view and click the restore icon on its row.
Cascade archive when Enders are linked
If the prompt is referenced by one or more Enders that are not already archived, a second Warning: Prompt in Use dialog appears after the first confirmation. The dialog:
lists each affected automation by name (click an entry to open the Ender in a new tab)
includes a warning that archiving the prompt will also archive every linked automation
Click Archive Both to archive the prompt and cascade-archive every linked Ender in a single step. Click Cancel to leave everything unchanged. The list of linked Enders is supplied by the backend, so it always reflects the current state of your automations.
Restoring is not cascaded
Restoring a prompt only unarchives the prompt itself. Enders that were archived as part of the cascade are not restored automatically — you must unarchive each affected Ender from the Enders list separately.
Enders That Reference Archived Prompts
When an Ender action is configured with a prompt that is later archived, the Ender cannot be enabled. If you toggle it on from the Enders list, the activation is reverted and a Cannot Enable Automation dialog appears listing every archived prompt the Ender depends on. To re-enable the Ender, edit it and switch each affected action to an active prompt, or restore the archived prompt first.
In Ender action editors, archived prompts still appear in the prompt picker but are shown with an Archived chip and cannot be selected.
Deleting a Prompt
Only archived prompts can be deleted. Switch to the Archived view and click the delete icon on the prompt row. Confirm the deletion when prompted. Deletion is permanent.