Prompt variables and tuning reference

Variables add saved settings and conversation details to a prompt. Nuvoca fills them with the available values during the call. Choose variables from the list in the editor you are using.

Agent prompt variables

Variable	Meaning
{persona}	Text from the Agent or Flow Persona field
{agent_name}	Configured Agent name
{important_note}	Additional note from the Agent settings
{conversation_history}	Previous turns, excluding the latest caller input
{conversation_fact_summary}	Facts explicitly stored for this conversation
{caller_phone_number}	Caller number, if available
{current_local_datetime}	Current local date, time and IANA timezone for the active configuration
{default_language}	Primary Azure recognition language; with Kugel derived from the primary selected language; with Deepgram de-DE
{detected_spoken_language_locale}	Most recently detected caller locale, empty before detection
{context}	Knowledge retrieved for the current turn
{query_transformation_context}	Information about the query transformation when applied
{pre_call_context}	Context from the configured pre-processing lookup

Place {persona} at the beginning of the main prompt, as shown in the prompt guide. Set greetings in the Agent settings and assign tools to the Agent. Neither needs an extra placeholder in the prompt.

Local time helps interpret phrases such as “tomorrow”. Clarify the caller’s timezone when arranging appointments if needed. The default language is a setting, while the detected language refers to the current caller.

Access individual caller-context fields

Use dot notation to read specific values from the processed pre-processing result. For example, {pre_call_context.customer.name} inserts the name from the customer object. {pre_call_context.orders[0].status} reads the status of the first list item. Indexes start at 0.

The path follows the stored output keys. If a mapping puts the name directly under name, use {pre_call_context.name}. See caller lookup for examples and missing-value behavior.

You can use these field paths in the Agent prompt, summary prompt and personalized greetings.

Routing variables

Routing prompts offer {conversation_history}, {conversation_fact_summary} and {previous_intent}. The previous intent is empty on the first turn. Use it to preserve continuity, while allowing a clear new caller request to change the route.

Missing values

Use the caller number only when present and confirmed by the caller.
If the lookup context is empty, continue with the standard questions.
If no retrieved answer supports a claim, say that the information is unavailable.

Copy variable names exactly from the editor. Connect the Knowledge Base or caller lookup before using its values in the prompt.

Find the right setting

Problem	Setting or guide
Repeated questions	Conversation history and prompt
Truncated responses	Model-specific completion budget
Missing or noisy facts	Search settings
Incorrectly recognized words	Recognition settings
Incorrectly spoken words	Pronunciation rules
Wrong specialist selected	Flow routing

Change one setting at a time and check the result with a new call.