Connect data: Tables, RAG, Memory

By the end of this page you will know how to connect your Task Agent to the three data source types available in the Data Vault: Tables, Content Stores (RAG), and Memory.

Prerequisites

A Task Agent created (Create a Task Agent).
Admin or Developer role.
Familiarity with the canvas (Canvas tour).

Understanding the Data Vault

The Data Vault is the project-level data layer accessible from the sidebar under your project. It contains four sub-sections:

Section	Purpose
Tables	Structured data storage — create tables, define schemas, insert/query/update/delete rows
Documents	Upload and manage files (PDFs, text, images)
Content Store	RAG-enabled knowledge bases — documents are chunked, embedded, and searchable via semantic search
Memory	Persistent key-value memory stores for agent context across conversations

Data Vault expanded in the project sidebar showing Tables, Documents, Content Store, and Memory sections

Connecting Tables

Tables let your agent read and write structured data. The canvas provides four Data nodes for table operations.

Create a table in Data Vault

Navigate to Data Vault → Tables in the project sidebar. Click Create Table, give it a name (e.g., “support_tickets”), and define columns with types (text, number, boolean, date, JSON).

Create Table dialog showing column name and type fields

Add a Data node to your agent

Open your Task Agent in the Agent Builder. From the Node Palette, expand the Data category. Drag one of the four available nodes onto the canvas:

Node	Description
Query Table	Query data from database tables with filters and sorting
Insert Row	Insert data into database tables with conflict handling
Update Row	Update records in a database table
Delete Row	Delete records from a database table

Configure the Data node

Click the node to open its configuration panel. Select the target table from the dropdown, then configure:

Query Table — Define filters (column, operator, value), sorting, and limit. Map the output to a variable.
Insert Row — Map input variables or static values to table columns. Configure conflict handling (skip, update, or error).
Update Row — Set filter conditions to match rows, then define the column values to update.
Delete Row — Set filter conditions to identify which rows to remove.

Query Table node configuration panel with table selection, filters, and output mapping

💡

Use variables from earlier nodes (e.g., output from a Quick Prompt node) as filter values. Reference them with the {{variableName}} syntax.

Connect the node in your flow

Wire the Data node between other nodes in your workflow. The output variable from a Query Table node is available to all downstream nodes.

Connecting Content Stores (RAG)

Content Stores enable semantic search over uploaded documents. Your agent can retrieve relevant knowledge chunks at runtime.

Create a Content Store

Navigate to Data Vault → Content Store. Click Create Content Store, name it, and configure chunking settings (chunk size, overlap). Upload documents or connect to external sources (DMS links, web crawlers).

Content Store creation interface with chunking configuration

Add a RAG node to your agent

From the Node Palette, expand the RAG category. Two nodes are available:

Node	Description
Search Content Store	Semantic search across a RAG content store — returns matching chunks
Reindex Content Store	Trigger reindexing of content stores, DMS links, or web crawlers

Drag a Search Content Store node onto the canvas.

Configure the Search Content Store node

Select the target Content Store, configure the search query (usually a variable from the Start node or a Quick Prompt output), set the number of results to return (top-k), and map the output to a variable.

Search Content Store node configuration showing content store selection and query settings

ℹ️

The Reindex Content Store node is useful in workflows that ingest new data and need to refresh the index before querying.

Connecting Memory

Memory stores provide persistent context across agent conversations — useful for remembering user preferences, session state, or accumulated knowledge.

Create a Memory store

Navigate to Data Vault → Memory. Create a new memory store with a name and optional description.

Use Memory in your agent

Memory is typically accessed through nodes that read/write key-value pairs. Configure a node to store output from your agent’s workflow or retrieve previously saved context at the start of a new execution.

💡

Memory is ideal for agents that need conversation continuity — e.g., remembering a customer’s previous requests across multiple interactions.

Data flow in the agent graph

Data moves through your agent via variables. Each Data or RAG node produces an output variable that downstream nodes can reference. The general pattern is:

Start node receives input variables (e.g., a user query).
Search Content Store uses the query to retrieve relevant context.
Quick Prompt combines the retrieved context with the user query to produce an answer.
End node returns the final output.

Common issues

“Table not found” — Ensure the table exists in the Data Vault and you’ve selected the correct one in the node configuration.
“No results returned” — Check your filter conditions. Try broadening the query or removing filters to verify data exists.
“Content Store not indexed” — New documents need indexing before they’re searchable. Wait for indexing to complete or trigger a manual reindex.
“Variable not available” — A variable is only accessible by nodes that are downstream of the node that produces it.