RAG Retriever Tool

Use RAG Retriever Tool to enable your agents to interact with external knowledge base(that is, Vector database) using predefined embedding models. This tool simplifies the process of accessing the external knowledge base to answer specific user queries by retrieving the top matches to answer the user query via embedding similarity.

Field Description
*Indicates mandatory field
Name*

A unique name to identify this instance of the rag_tool. Used to reference it in your agent’s logic.
Description* A brief explanation of this tool’s purpose in the agent’s workflow (for example, A tool that uses a retriever to answer questions).
Vector Dimension An Interger (for example, 768) Value to provide for initializing the OpenAIEmbeddings. This number must exactly match the vector dimension of the embedding model you used to create your vector database. If this number is incorrect, the search will fail.
Top Match This specifies the number of relevant documents (or text "chunks") to retrieve from the database.(for example, 3).
Embedding Model Name

This defines the model that will be used to convert the user's live query into a vector for searching.

This must be the exact same model that was used to create the vectors in your database. A mismatch will result in poor or irrelevant search results.
Credentials for Database Type*

This section connects the tool to your database and the embedding model service.
Database Type* Select the type of vector database you are using (for example, PGVector, PGVector Store and so on).
Credentials to connect with*

From the dropdown, choose the pre-configured credential (like an API key, connection URL, or collection name) that allows the tool to access your database.

Must be pre-configured in the Credentials section.

Note: Refer to the Adding a Credential section for steps to configure this.
Credentials for Service provider*
Service Provider* Select the service that hosts the embedding model (for example, Hugging Face, OpenAI, a private endpoint).
Credentials to connect with* Choose the pre-configured credential (like a Hugging Face API, OpenAI token) required to use the embedding model specified earlier.