Knowledge

Important: Agents is currently in private beta and is not yet available to all customers.

Use Knowledge to provide information sources that the agent can reference when generating responses. These sources allow the agent to retrieve relevant information during execution.

You can add knowledge from document stores, vector embeddings, or outputs from earlier agents in the flow.

Jump to:

Shared context
Document store
Vector embedding

Shared context

Enable Shared context to allow the agent to access outputs from earlier agents in the same flow.

Field or selection	Description
Shared context	Allows the agent to access outputs generated by earlier agents in the flow. Enabling this option increases token usage.

Document store

Use Document store to select document repositories that contain information the agent can reference. The agent searches the selected document store to retrieve relevant documents when generating responses.

Field or selection	Description
Document store	Select the document store that contains documents the agent can reference.
Describe knowledge	Provide a description of the knowledge base. This helps the AI understand the type of information available and when to search this source.
Return source documents	Specify whether the agent should return the source documents used to generate the response.
Add document store	Adds another document store.

Vector embedding

Use Vector embedding to configure knowledge sources based on vector embeddings. Vector embeddings allow the agent to search indexed content and retrieve relevant information based on semantic similarity.

Field or selection	Description
Vector stores	The vector store that contains the embeddings the agent can use. For more details, see Vector stores.
Embeddings	The embeddings available in the selected vector store.
Knowledge name	A short name for the knowledge base. This helps the AI determine when to use this knowledge source.
Knowledge description	Provide a description of the knowledge base. This helps the AI understand the type of information available and when to search it.
Return source documents	Specify whether the agent should return the source documents used to generate the response.
Add vector embedding	Adds another vector embedding source.

Vector stores

Vector store	Fields
Chroma	Open-source embedding database designed for AI applications. Stores vector embeddings and supports similarity search to retrieve relevant documents. Connection: The connection used to access the Chroma instance. Collection name: Name of the collection that contains the embeddings. (Text, Decimal, Integer, Boolean, DateTime, Collection)
MongoDB Atlas	Managed cloud database service that supports vector search through MongoDB Atlas Vector Search. Connection: The connection used to access MongoDB Atlas. Database: Name of the database that contains the vector collection. Collection name: Name of the collection that stores the embeddings. (Text, Decimal, Integer, Boolean, DateTime, Collection) Index name: Name of the index used for vector search. (Text, Decimal, Integer, Boolean, DateTime, Collection)
OpenSearch	Distributed search and analytics engine that supports vector search for retrieving semantically similar data. Connection: The connection used to access the OpenSearch instance. Index name: Name of the index that stores the embeddings. (Text, Decimal, Integer, Boolean, DateTime, Collection)
Pinecone	Managed vector database designed for large-scale similarity search and AI applications. Connection: The connection used to access Pinecone. Pinecone index: Name of the Pinecone index that stores the embeddings. (Text, Decimal, Integer, Boolean, DateTime, Collection)
Postgres	Relational database that supports vector search using extensions such as pgvector. Connection: Select the connection used to access the PostgreSQL instance. Host: Host address of the database server. Database: Name of the database that stores the embeddings.
Qdrant	Open-source vector database designed for high-performance similarity search. Connection: Select the connection used to access the Qdrant server. Qdrant server URL: URL of the Qdrant server. Qdrant collection name: Name of the collection that stores the embeddings. Vector dimension: Size of the embedding vector. Similarity: Similarity measure used to compare vectors.
Redis	In-memory data store that supports vector similarity search. Connection: Select the connection used to access the Redis instance. Index name: Name of the index used for vector search. Replace index on upsert: Specifies whether the index is recreated when upserting data.