2025.19.0 - October 1 2025 Preview Release

Changes are scheduled to go to production on October 15 2025

New Features

AI Discovery Improvements

• Added Rerank Guidance Prompt to the AI Discovery Experience - a new prompt in addition to the existing Rephrase and Confidence prompts. This prompt is intended to instruct the AI agent to order the results in the described order. For example, asking the AI to ensure it orders the results by price in increasing order if the user query alluded to cost being a consideration. The addition of a Rerank prompt comes at the cost of longer response times, as the AI Agent will need additional time to process the results with the Rerank prompt (if one exists in the experience).

• Added the option to choose between different LLMs for the AI Agent, as well as separately for generating the vector data. Currently this only includes the option of the OpenAI or Google LLMs for the AI Agent. The generation of the vector data can only be done with the OpenAI LLM at the moment. The default is set automatically to the OpenAI GPT 4.0 mini LLM. One reason you might want to change to the Google LLM is to reduce response processing times if the OpenAI LLM proves to be too slow. There is no guarantee that Google's Gemini LLM will be any faster. To request a change to the Google Gemini 2.0 Flash please open a ticket by emailing [email protected]

Improvements & Bug Fixes

• Fixed the AI Discovery feature to allow empty rephrase and confidence prompts. In case the confidence prompt is left empty, the platform simply uses a default confidence prompt on the backend instead. In case the rephrase or the new rerank prompts are left empty, then the AI will skip evaluating those steps altogether.

• Fixes to the /api/assets/export endpoint to increase stability of execution

• Improved HTTP security headers for web application, embeds, and rendering