Generative Search

policies:
# - ...
  - name: EnterpriseSearchPolicy
    llm:
      model: "gpt-4.1-mini-2025-04-14"
# - ...

{% if check_relevancy %}Based on the provided documents and the recent conversation context, answer the following question.
Before responding, ensure the answer is directly supported by the documents or context.
Do not make assumptions or infer beyond the given information.
Only answer if you are more than 80% confident that the response is fully supported.
If the answer cannot be determined, respond with: [NO_RAG_ANSWER]
{% else %}Given the following information, please provide an answer based on the provided documents and the context of the recent conversation.
If the answer is not known or cannot be determined from the provided documents or context, please state that you do not know to the user.
{% endif %}

### Relevant Documents
Use the following documents to answer the question:
{% for doc in docs %}
{{ loop.cycle("*")}}. {{ doc.metadata }}
{{ doc.text }}
{% endfor %}

{% if citation_enabled %}
### Citing Sources
Find the sources from the documents that are most relevant to answering the question.
The sources must be extracted from the given document metadata source property and not from the conversation context.
If there are no relevant sources, write "No relevant sources" instead.

For each source you cite, follow a 1-based numbering system for citations.
Start with [1] for the first source you refer to, regardless of its index in the provided list of documents.
If you cite another source, use the next number in sequence, [2], and so on.
Ensure each source is only assigned one number, even if referenced multiple times.
If you refer back to a previously cited source, use its originally assigned number.

For example, if you first cite the third source in the list, refer to it as [1].
If you then cite the first source in the list, refer to it as [2].
If you mention the third source again, still refer to it as [1].

Don't say "According to Source [1]" when answering. Instead, make references to sources relevant to each section of the answer solely by adding the bracketed number at the end of the relevant sentence.
#### Formatting
First print the answer with in-text citations which follow a numbered order starting with index 1, then add the sources section.
The format of your overall answer must look like what's shown between the <example></example> tags.
Make sure to follow the formatting exactly and remove any line breaks or whitespaces between the answer and the Sources section.
<example>
You can use flows to model business logic in Rasa assistants. [1] You can use the Enterprise Search Policy to search vector stores for relevant knowledge base documents. [2]
Sources:
[1] https://rasa.com/docs/rasa-pro/concepts/flows
[2] https://rasa.com/docs/rasa-pro/concepts/policies/enterprise-search-policy
</example>
{% endif %}

{% if slots|length > 0 %}
### Slots or Variables
Here are the variables of the currently active conversation which may be used to answer the question:
{% for slot in slots -%}
- name: {{ slot.name }}, value: {{ slot.value }}, type: {{ slot.type }}
{% endfor %}
{% endif %}
### Current Conversation
Transcript of the current conversation, use it to determine the context of the question:
{{ current_conversation }}

{% if current_datetime %}
### Date & Time Context
- Current date: {{ current_datetime.strftime("%d %B, %Y") }}   (DD Month, YYYY)
- Current time: {{ current_datetime.strftime("%H:%M:%S") }} ({{ current_datetime.tzname() }})  (HH:MM:SS, 24-hour format with timezone)
- Current day: {{ current_datetime.strftime("%A") }}     (Day of week)
{% endif %}

## Answering the Question
Based on the above sections, please formulate an answer to the question or request in the user's last message.
It is important that you ensure the answer is grounded in the provided documents and conversation context.
Avoid speculating or making assumptions beyond the given information and keep your answers short, 2 to 3 sentences at most.

{% if citation_enabled %}
If you are unable to find an answer in the given relevant documents, do not cite sources from elsewhere in the conversation context.
{% endif %}

Your answer: