Assistant Prompts

This widget is designed to configure how TellusR assistants process user queries and generate responses. It allows you to fine-tune search behavior, response merging, and assistant setup to optimize interaction with your data.

TellusR supports two types of assistants:

  • Standard Assistants – Enable direct interaction with a search project, allowing users to chat with content from a specific dataset.
  • Stitch Assistants – Combine responses from multiple assistants into a single reply, either by stitching them together as separate answers or by merging them into a unified response using an LLM.

Assistant Configuration

New assistants can be created by clicking the “+"-sign next to the assistant selection menu.

Assistant selection in menu

When creating an assistant, you can choose between a Standard or Stitch assistant, depending on your needs.

Widget Settings

Standard Assistant Settings

  • Search Project: Selects the dataset the assistant will use for search-based chat.
  • Top N: Specifies the number of top-matching chunks used for generating responses (excluding neighboring chunks).
  • Semantic Weight: Controls the balance between semantic and keyword-based search (0 = keyword-based, 1 = semantic).
  • Highlight Window: Determines how many neighboring chunks are included for additional context in responses.
  • Query Parsing Mode:
    • Normal: Generates search query suggestion from user input.
    • Stable (Blend Three): Asks the LLM three times for query suggestions and blends the results.
Standard assistant settings

Stitch Assistant Settings

  • Stitch Profiles: Defines which assistants are included in the stitch assistant.
  • Merge Stitch Reply Prompt: When LLM Merge = true, this prompt guides how the LLM merges responses while keeping references intact.
  • Enable LLM Merge of Stitch Reply:
    • False → Replies from multiple assistants are combined as separate answers.
    • True → An LLM merges responses into a single, cohesive reply.
Stitch assistant settings