Chat with an AI Agent

Video walkthrough

What this does

Chat is the main way to talk to a configured AI agent in Bruin Cloud. Depending on how the agent is set up, you can use it as an AI data analyst that queries your warehouse, or as an assistant for data engineering tasks like checking pipeline status and run history.

If you haven't created an agent yet, see Configure AI Agents first.

Steps

1) Open chat

From the AI menu, navigate to Chats.

2) Pick an agent

Use the dropdown to select the agent you want to talk to. Each agent comes with the project, connection set, integrations, and system prompt you configured for it.

3) Send a prompt

Type your question in the message box and send it.

A useful first prompt is to ask the agent what it has access to and what kind of data you can ask about. This makes it explore its environment and surface what's available before you dig in.

4) How the agent works under the hood

When you send a prompt, the agent:

• Spins up a sandbox environment.
• Clones the repo of the project it's connected to.
• Reads your pipelines, assets, and any AGENTS.md or instruction files.
• Builds a memory of that context so it can answer questions in the right scope.
• Uses the agent's connection set to query the data warehouse - typically starting by inspecting the schema and mapping out available tables.

Example - generate a financial report

In the video, the agent has access to a warehouse with sample stock-market, weather, and other datasets.

The prompt: "Create a financial report for Microsoft and Apple as a PDF."

From the prompt and the warehouse metadata, the agent:

• Identifies which datasets and tables to query.
• Runs the queries it needs.
• Generates charts.
• Attaches both a Python file and a PDF for download.

In the example, this took about 23 steps and 6 queries.

Example - operate Bruin Cloud via CLI

If the agent has Cloud CLI access enabled, you can also use it for data engineering tasks. It can:

• Read pipeline run history
• Inspect assets and the catalog
• Run pipelines and check pipeline status

Sample prompt: "How many times has pipeline X run in the last 10 days, and how many of those failed?"

In the video, the agent worked through 11 steps in about a minute and reported back: 7 runs in the last 10 days, 3 of them failed.

Key takeaways

• Pick the agent in the chat dropdown - its access (project, connection set, CLI) is set on the agent itself.
• Ask the agent what it has access to before diving into specific questions.
• Data-analyst use cases - schema exploration, queries, charts, PDF reports.
• Engineering use cases - pipeline status, run history, catalog lookups, triggering runs.
• Each chat runs in a sandbox that clones the connected repo and builds context from your AGENTS.md files.

Next

Want to lock down what the agent can read or write? Set up a dedicated connection set and attach it from Configure AI Agents.