Add a retriever
Add a retriever to your prompt to augment and add context to your prompts with AI search results.
Before you begin
Role required: sn_skill_builder.admin
About this task
Using a retriever in a skill enhances the relevance and coherence of a response by pulling relevant information from a source and then feeding it to a large language model (LLM) to create the final output. This process leads to smoother conversational flow and better scalability as data grows, without requiring model fine-tuning.
A retriever enables the chatbot to access external knowledge by fetching relevant background information, resulting in more factual, in-depth, and informed responses.
Procedure
Navigate to All > Now Assist Skill Kit > Home.
Create a skill or select the skill that you want to add a retriever to.
Select the 2. Add tools tab.
Select (+) icon to add a node.
Select Tool node.
Select Retriever from the drop-down menu.
Select Configure retriever.
On the form, fill in the fields.
| Field | Description |
|---|---|
| Name | The name for the retriever. |
| Search query | The information that you want to search for. It can be static text or a skill input. |
| Search space type | - Table-based - Search-profile-based |
| Search profile | A group of search sources. |
| Search sources | Tables in ServiceNow that have been indexed and can be used for search. |
| Fields returned | The fields that you want returned from the search sources and sent to the large language model \(LLM\). |
| Limit | The maximum number of results that are returned. |
| Search criteria |
|
Select Next.
Select an embedding model.
To learn about embedding models, see .
If you selected Hybrid or Semantic search criteria, select a semantic index.
A semantic index enables you to search for data based on contextual meaning.
If you selected Hybrid or Semantic search criteria, select Advanced to change the document matching threshold.
The document matching threshold is the threshold for semantic matching. The default value is 0. You can input any value from 0 to 1. You can get more precise results with a higher value.
Select Next.
Note: If you selected Hybrid or Semantic search criteria, see Retriever chunking and reranking to complete setting up your retriever.
Review the retriever tool information.
Select Add tool.
Retriever chunking and reranking
When you’re building a skill prompt that uses a retriever you can use chunking and reranking to enhance the accuracy and relevance of your responses.
Parent Topic:Create a prompt