AI agents

AI agents add a chat assistant to your Power BI reports. Instead of clicking through visuals or building a new report, your users just ask a question in plain language, like “how many hours did we log per project last year?”, and get an answer back with the numbers, a table, or a chart. The agent runs the query against the report’s semantic model itself, so every answer is based on your real data, not a guess.

Because it queries the model directly, nobody needs to know DAX or prepare anything up front. Ask a follow-up, change the filter, or compare two periods. It’s a conversation, not a report you have to design.

A few things worth knowing before you start:

• It runs on your own AI key. You bring your own OpenAI or Anthropic key, so the model, the cost, and the data stay under your control. Nothing runs through a shared DataTako key.
• It respects row-level security. Each user only gets answers about the data they’re allowed to see. The agent queries under their RLS, exactly like the report does.
• You attach it per report. An agent is tied to a report’s dataset, so its answers are always scoped to that model.
• You stay in control of usage. Set budgets and limits per agent, and follow spend and usage so there are no surprises.

The guide below walks you through it step by step: adding your API key, creating an agent, and attaching it to a report.

To create your first AI agent, navigate to the AI Agents menu.

Before we can add an agent, we first need to add an API key. DataTako works with the ‘bring your own API key setup’. This means you can bring your own AI model API key, we support the following sources:

• Anthropic (Claude models)
• OpenAI
• Azure AI
• Azure Foundry

In this article we assume you already know how to obtain an API key for one of the supported providers above. Missing a provider? Let us know and we’ll most likely are able to accommodate in an upcoming release!

API keys

Navigate to the API keys tab and click on the ‘Add API key’ button. You’ll see this window:

From the provider list, select the provider you want to add an API key from.

After adding an API key, switch back to the AI agents tab.

AI agents

Click on the ‘Add AI agent’ button, in the window that opens, configure the agent:

API key (required)
The provider key the agent runs on. Pick one you added earlier on the API keys tab, or click Add new API key to add one now. This is your own OpenAI or Anthropic key, so usage is billed to your provider account and not through DataTako.

Model (required)
The specific model to use, loaded from the key you selected. You have to pick an API key first before this list fills. Pick a current, capable model. A stronger model gives noticeably better answers on real questions. We recommend selecting one the models that is marked with (recommended) in the dropdown.

Instructions (optional)
Free-text context for the agent: a persona (“answer like a friendly analyst”), business definitions, and synonyms your users use for the same thing (“turnover means revenue”). This doesn’t open up any data. The platform’s security and guardrails always apply, and you connect the agent to an actual dataset later, in the report configuration screen.

Reliable answers (recommended)
Leave this on. It routes everyday questions through a more predictable method that returns steadier, more accurate answers. Harder questions fall back to a different approach automatically, so you lose nothing by keeping it on.

Extra thorough for complex questions (experimental)
For hard questions, the assistant retries a few times and checks its own result until it holds up. It’s more accurate on tricky asks but slower and more expensive per question. You usually don’t need it when Reliable answers is on, so turn it on only if you see complex questions coming up short.

Example questions (optional)

This is the single biggest lever on answer quality. Add a few questions your users actually ask, each paired with a DAX query you know returns the right result. The agent uses these as worked examples and answers similar questions more accurately. Click Add example for each one.

Even three or four good examples make a clear difference. Focus on the questions that matter most, or the ones the agent tends to get wrong.

When you’re done, click Save.

Add agent to report

Open a report configuration and navigate to the AI assistant tab. In this tab, select the previously configured agent:

After enabling the AI assistant for this report, when you open the report, it now has an ‘AI’ floating button at the bottom right:

Clicking on it opens the AI page at the right side, similar to the report comments feature:

In this chat, you are able to ask questions about the data, ask for all kind of details, the sky is the limit!

Add standalone agent

It’s also possible to add a standalone chat agent without having to open the report, it features a full page chat page.

In the report configuration screen, click on the ‘AI assistant (chat only)’ button:

The configuration is somewhat similar to the regular report configuration:

Instead of selecting a report to share, you select a semantic model that the AI agent is able to use. After configuration, the AI assistant shows up like a regular report in the overview.

After opening the assistant from the overview, you’ll see something like this:

Chat history and multiple chats

Your chat history is saved, you can also start new chats easily with the buttons at the top right of the chat window (supported by both stand-alone and in-report):