Prompt expansion

You may use handlebars-like branching logic in prompt:

{{#if variable == "value1"}}
prompt-element-1
{{#elseif variable == "value2"}}
prompt-element-2
{{#else}}
prompt-element-3
{{/if}

Conditions syntax

Conditions syntax is very flexible and supports the following elements:

• Comparison operators

• Type casting

• Logical operators

• Brackets

• String operators

• Array operators

• Nesting

Comparison operators

You may use the following comparison operators: ==, !=, <, >, <=, >= .

For variables that contain string values, enclose the values that you compare with in quotes. For example:

{{#if sex == "male"}}

You may also test for empty string by simply specifying the variable name:

{{#if customer_name}}

Variables are automatically assigned one of the following four types – string, integer, float, boolean – based on the provided value. This allows you to perform numeric comparison, for example:

{{#if age > 18}}

Make sure that variable type always matches the variable type.

Type casting

You may explicitly cast variables to specific type using str(), int(), float(), or bool() methods.

For example:

{{#if str(age) == "18"}}

Logical operators

You may use and, or, not logical operators in your conditions.

For example:

{{#if user_utterance == "male" and age > 18}}

Brackets

You may use brackets to group logical condition or specify multi-line conditions.

For example:

{{#if (person_name == "John" or
           person_name == "Jack" or
           person_name == "Jonathan")}}
Say "Hello, Mr. J!"
{{/if}}

String operators

You may use the following operators on variables of type string: .lower(), .upper(), .strip(), .lstrip(), .rstrip(), .startswith(prefix), .endswith(prefix)

For example:

{{#if person_name.startswith("J")}} Say "Hello, Mr. J" {{/if}}

You may also use in and not in operators to perform partial string comparison.

For example:

{{#if "please leave a message" in user_utterance}} End the call {{/if}}

Array operators

You may use in and not in operators to compare values against arrays. The latter should be specified as comma-delimited list enclosed in brackets.

For example:

{{#if person_name in ("John", "Jack", "Jonathan")}} Say "Hello, Mr. J" {{/if}}

Nesting

You may use nested #if / #elseif constructs to create complex conditions.

Expansion process

It is important to understand that expansion happens before the prompt is sent to the LLM. The following examples illustrate this concept.

Example 1

For prompt:

{{#if role == "admin"}}
User has full access to the system.
{{#elseif role == "user"}}
User has limited access to the system.
{{#else}}
User is not allowed to access to the system.
{{/if}}

Assuming that the role variable is set to admin, LLM will receive the following:
User has full access to the system.

Example 2

For prompt:

{{#if user_age > 18 and subscription_status == "active"}}
User has access to premium content
{{#else}
User doesn't have access to premium content.
{{/if}}

Assuming that the user_age variable is set 16, LLM will receive the following:
User doesn't have access to premium content.

Example 3

For prompt:

Hello {{#if gender == "male"}}Mr{{#else}}Mrs{{/if}} Smith!

Assuming that the gender variable is set male, LLM will receive the following:
Hello Mr Smith!

Use of time variables in conditions

If you want to use time variables in #if / #elseif conditions, you must use the following special syntax: "{current_time}". For example, enclose them in curly brackets and quotes. Time variables get expanded before the rest of the prompt is evaluated, and this special syntax ensures that they are treated as regular strings during conditions evaluation.

For example:

Say good {{#if "{current_time}" < "10:00"}} morning {{#else}} day {{/if}}

Dynamic prompts

By default, prompt expansion happens at the beginning of the conversation, as part of agent’s initialization.

Alternatively, you can use dynamic_prompt advanced configuration variable to re-generate prompt for specific agent dynamically, on each user utterance.

When dynamic prompts are enabled, the following additional variables are provided by the Live Hub platform:

Variable	Type	Description
user_utterance	string	Last user utterance
user_utterance_count	integer	Starts with zero and is incremented with every user utterance
user_message_count	integer	Similar to user_utterance_count, but counts only real message said by user, and ignores, for example, NO-USER-INPUT utterances.
duration	integer	Duration of the call in seconds

The following example demonstrates the use of dynamic prompts in an outbound dialer agent to ensure that the call is answered by a real human, and not IVR’s or answering machine.

{{#if ("your call is important" in user_utterance or
       "stay on the line" in user_utterance or
       "wait while we connect" in user_utterance or
       "NO-USER-INPUT" in user_utterance)}}
    {{#if duration < 40}}
        Say "Waiting for human to join"
    {{#else}}
        End the call
    {{/if}}
{{#else}}
    Introduce yourself.
    Inform user that his car was evacuated.
    ...
{{/if}}