System Prompt
The system prompt allows you to control the behaviour of the model. You can input variables using placeholders, and conditional text using JINJA Python syntax.
A system prompt is created each time you post a prompt back to a model, even mid-conversation.
Available System Prompt Variables
currentDatetime - The current date and time in format YYYY-MM-DD HH:mm:ss.
username - The user's login username.
givenName - The user's firse name.
familyName - The user's last name.
email - The user's email address.
departmentName - The user's department.
jobTitle - The user's job title.
text - The combined text of all the documents in the search results.
percentageTokensUsed - The percentage of tokens that have been used for the context window (results).
percentageTokensLeft - The percentage of tokens that are left for the context window (results).
hasTermsFilters - True if the search has any terms filters, false otherwise.
chatCount - The number of chat messages in the chat history.
filtersChanged - True if the filters have changed since the last chat message.
The below image shoes a more complex agentic system prompt that will perform a deep research task. It calls tools several times to compile a response for a user.
Last updated