Grok Bot Data Bot: what it is and how to set it up
2026-11-08 · 5 min read
Data Bot is a shared agent inside Grok Bot that queries a company's data warehouse and returns answers, charts and written reports to anyone on the team. To set one up you need a Teams or Enterprise plan, then you add the prebuilt version or create your own, connect read-only data access, and teach it your metrics.
The posts behind it
On November 2, @bot, the official Grok Bot account, said its team built something that "gives everyone at the company on-demand access to a data scientist." Product managers use it to look into product usage, and marketers use it to measure campaigns.
A follow-up post put a number on it: Data Bot "handled an estimated 70 hours of work" in the prior week that would otherwise have gone to the data scientists. That figure is the company's own estimate and nobody outside has checked it.
xAI, whose site now carries the SpaceXSI name, explained the build in its write-up. Data Bot runs as a Team Bot, a Bot shared across a company, and it's built on the same Grok Bot agent we covered when it learned to make calls.
What it can do
The data team teaches it through shared skills in four areas: metrics and analysis, experiments, charts and dashboards, and data engineering. The metrics skill is the one I'd copy. It records each core number with its source table and query logic, so two people asking about revenue get the same revenue.
These are xAI's examples, reworded as prompts you could type:
| Who asks | Prompt | What comes back |
|---|---|---|
| Marketer | "Of the people who clicked this campaign, how many switched plans?" | A count the email tool's own conversion metric couldn't show |
| Product manager | "Why did adoption of this feature change last month?" | A planned analysis and written report with charts or a Hex dashboard |
| Revenue lead | "How are we tracking against this week's target?" | Progress numbers and charts for a presentation |
| Whole team | "Post this metric here every Monday." | A scheduled Slack update people can question in the thread |
When a number spikes or drops, it's supposed to break the change down by segment or region and check for holidays and logging changes before blaming the product. I'd ask the same of a human analyst.
How to set up Data Bot
Team Bots are "available today in public beta on Teams and Enterprise plans," per the September announcement. A Premium X Subscription gets you a personal Grok Bot, but a shared Bot needs a team.
- Turn on Grok Bot. On Teams it's on by default. On Enterprise, an admin enables it from the dashboard.
- Add the prebuilt Data Bot with the "Add to Grok Bot" link on the announcement page, or create your own: click "+" in the sidebar, choose "Create a Bot," name it, and tell it the job.
- Connect data with a read-only login limited to approved tables. xAI's own version uses shared, read-only credentials on Databricks, plus Datadog, Hex and Statsig.
- Write the skills: each metric's definition, source table and query logic, and the pitfalls your data team already knows about.
- Click the Bot's name, then "Share." Pick your team or specific people, and make each one a Collaborator (chat only) or a Manager (can change skills, memory and apps).
- Click "Bring to your team's Slack," then "Connect." A Slack admin may need to approve it.
Expect step 4 to take the longest. i-scoop reports that customers get a skeleton instead of xAI's internal skills library, so you'd have to do the teaching yourself before seeing anything like 70 hours. I couldn't find a published Team Bot price, and xAI's page sends buyers to sales.
Check these before you connect anything
A shared Bot uses the apps and secrets its owner added for everyone who chats with it, so a colleague can reach whatever the owner connected. Set it up with a separate service account instead of your own login.
Vault, Grok Bot's built-in login store, isn't available on shared Bots. If you read our post on how Grok Bot handles agent logins, don't assume that setup carries over.
The same security page says Grok Bot computers run in the United States and customer data isn't used for training with Privacy Mode on. It also advises starting new roles on read-only tasks and draft outputs.
My read
I like the review step most. Data Bot can propose things to remember, such as a metric drop that turned out to be a regional holiday, and managers "approve or reject them before Data Bot shares that knowledge with everyone who uses it." One wrong fact repeated to a whole company does real damage, and making a person sign off costs very little.
xAI's version also had a data science team writing the skills, which most buyers won't have. If nobody can say how your numbers are defined, expect confident answers built on loose definitions.
What this means for businesses
Most small companies don't run Databricks, but the recipe scales down: pick one question people keep asking, write down exactly how that number is calculated, give the agent read-only access, and have a person approve what it learns. If you'd like help finding that first question in your own operation, New Face Design offers a free process audit.