Ground the Agent with an Index
Add semantic retrieval so the agent can find and use relevant organizational data.
Ground the Agent with an Index
An agent can use a connection such as Google Drive directly, without an Index. That works well for targeted actions or live lookups. An Index creates a semantic, searchable copy optimized for fast retrieval across many documents and records.
Indexes support document question answering and deeper research over organizational data such as call transcripts, support records, and files from Google Drive.
How indexed data works
A Data Loader imports each integration record as a BotDojo document. The document includes its content and metadata: a JSON object of attributes and related information about the record. During indexing, BotDojo splits the content into searchable parts and creates embeddings.
When an agent searches, BotDojo combines semantic similarity with metadata filtering. That lets the agent find relevant meaning while narrowing results by attributes such as source, account, date, or record type. For large indexes, materialized columns extract frequently queried metadata into dedicated columns; column indexes can optimize those filters further.
Add Search Index as an agent tool
First, create and populate an Index. Then return to the Flow Designer:
- Select Add Node and search for
Index. - Under Agent Tools, drag Search Index onto the canvas.
- Configure the tool to use your Index.
- Connect the Search Index tool output to the AI Agent's Tools input.
Search Index is the simplest tool for grounding this agent. Use Index MCP when an agent needs a broader set of index and metadata operations.
For reliable answers, tell the agent in its System Prompt when to search and how to respond when the Index does not contain enough evidence. See Metadata and Markdown to control searchable content and filterable attributes.