Data Chat User Guide

Table of Contents

1. Logging In
2. Starting a New Chat
3. Selecting Your Agent
4. Asking Questions
5. Understanding the Response
6. Building Visualizations & Dashboards
7. Giving Feedback

---

1. Logging In

To access the platform:

1. Open a Chrome browser and go to your platform's login page.
2. On the login screen you'll see three options:
   • Login with SSO
   • Login with password
   • Login with magic link
3. Enter your email address.
4. Click Login with password.
5. Enter your password.
6. (Optional) Allow Chrome to save your login details so you don't have to re-enter them each time.

Once authenticated, you'll be taken to the main chat interface.

---

2. Starting a New Chat

• Use the left-hand sidebar to navigate the app.
• Click the New Chat icon to open a fresh conversation.
• Your previous conversations appear in the sidebar under Chats, grouped by date (e.g. "Today"), so you can return to any earlier session.

---

3. Selecting Your Agent

Before asking a question, choose the right model or agent:

1. Click the model/agent selector at the top of the chat window (it shows the current model, e.g. claude-opus-5).
2. From the dropdown you can browse:
   • Anthropic models
   • My Agents (your custom, data-connected agents)
   • Other providers (e.g. Azure OpenAI)
3. Under My Agents, select your data agent (for example, a "Data Agent" connected to your organization's warehouse).
4. Once selected, you'll see a welcome screen confirming you're ready to chat with your data.

Tip: The data agent is connected to your live data source (via MCP servers), so it can answer questions using your actual, up-to-date figures.

---

4. Asking Questions

Type your question in natural language into the message box at the bottom of the screen and send it. You don't need to write SQL or know the underlying table structure — the agent translates your request for you.

Example questions you can ask:

• "Which region received the lowest score in July 2025?"
• "Which area had the highest monthly performance for [metric A] and [metric B] in [region], broken down by month for 2025?"
• "I want to see the performance metrics for 2025. Create a visualization for each one so I can understand what improved, what stalled, and what we need to look into."

Tips for better answers:

• Be specific about the time period (month, quarter, year).
• Name the metric, region, or dimension you care about.
• Ask for a breakdown (e.g. "by month," "by region") when you want detail.
• You can ask follow-up questions in the same chat — the agent keeps the context.

---

5. Understanding the Response

When the agent answers a data question, it typically shows:

• Thoughts — the agent's reasoning steps (expandable).
• Tools used — which data tools/sources it queried (e.g. your warehouse connector). You can expand this to see what ran.
• Interpreted question — a plain-language restatement of how the agent understood your request (which metrics, dimensions, filters, and time period it used). Always check this to confirm it matched your intent.
• Results — usually a clean table or summary answering your question directly.

If the interpretation doesn't match what you meant, simply clarify in a follow-up message and the agent will adjust.

---

6. Building Visualizations & Dashboards

The agent can turn your data into interactive charts and dashboards, not just tables.

• Ask it to "create a visualization" or "build a dashboard" for any metric or set of metrics.
• It can produce elements such as:
  • KPI summary cards (e.g. yearly average, number of regions above/below target).
  • Monthly trend charts highlighting months that dropped or improved.
  • Ranked bar charts showing biggest improvements and biggest declines.
  • Toggle views to switch between different metric groups.
• Dashboards are interactive — you can filter and explore them directly in the chat.

Tip: When you want an overview, ask for "what improved, what stalled, and what needs attention" — the agent will lay out the right mix of visuals automatically.

---

7. Giving Feedback

Your feedback helps improve the agent's responses.

1. Below any response you'll see feedback controls, including thumbs-up and thumbs-down buttons (thumb buttons).
2. Click thumbs-up if the answer was helpful, or thumbs-down if it wasn't.
3. Clicking thumbs-down a second time opens a "Provide additional feedback" text box, where you can:
   • Type a detailed note explaining what was wrong or what could be better.
   • Click Save to submit your feedback, or Delete to discard it.

This extra detail helps the team refine the agent over time.

---

Quick Reference

---