Skip to main content

Before you start

Make sure you have at least one BigQuery connector set up in Settings → Connectors. Your agent needs a connector to query live data.

Creating a new agent

1

Go to Agents

From the left sidebar, click Agents and then Create agent.
2

Set a name and description

Give your agent a clear, descriptive name — for example, “Paid Search Analyst” or “Campaign Budget Tracker”.
3

Choose an agent type

Select Data Analysis Agent for most use cases. Use Multi-Agent Swarm if you need separate, independently configured subagents. See General settings for a comparison.
4

Configure the Agent Builder

Write the orchestrator system prompt — tell the agent what it does, which data it works with, and how to format its responses. Agent Builder →
5

Set up data access

Connect a data source, restrict which datasets and tables the agent can query, and control which users can access it. Data Access →
6

Save and test

Click Save, then Open chat to test your agent. Adjust the system prompt and context until the results match your expectations.

Agent settings

Each agent has five configuration tabs. Click any card below to jump to the detailed reference.

General

Name, slug, agent type, model, and display settings.

Agent Builder

Orchestrator prompt, subagents, tools, model overrides, and context.

Data Access

Data source, connector, dataset and table filters, and user group restrictions.

Guardrails

Input and output checks: PII masking, off-topic detection, custom rules.

Instructions & UX

Agent logo, start-page instructions, and example prompts.

Testing your agent

Once saved, click Open chat to test your agent. Check:
  • Does it query the right tables?
  • Does it use the correct date ranges and filters?
  • Is the output format what you expected?
Each time you save, a new prompt version is created — you can always roll back to a previous version. Learn about prompt versioning →