AI-assistant

Two things need to be ready before you can ask anything: you must be signed in, and the in-browser R runtime that powers it has to finish starting up. That runtime starts loading as soon as the explore map does, so by the time you've found and clicked the icon it's usually ready — a “Starting assistant” state covers the gap otherwise.

Your question isn't tied to whatever point is currently selected on the map — name a place in the question itself (“Tokyo”, “Rome and Berlin”) and the assistant geocodes it independently.

1. 1The AI writes R code. Your question is sent off and comes back as R code — not a text answer — built to call the same historical data used everywhere else on the platform.
2. 2That code runs in your browser. The code runs through the same in-browser R engine as r-lab, right there in the tab — nothing is executed on a server, and no data leaves your session beyond the API calls the code itself makes.
3. 3The result renders inline. Depending on what the code produced, you get a chart, a table, a rendered plot, or plain text — whichever fits the data best.

Chartable results default to a chart, with a toggle to switch to the raw table. Results the assistant can't chart (like a base-R plot) render as an image instead. Every answer also has a View R code  section — the exact code that ran, rewritten to swap in your real API key so it's a standalone script you can copy and run outside the platform, e.g. in RStudio.

Every finished answer has a Feedback button — say whether it answered your question and optionally add a note. That's sent along with the question and the exact code that ran, so a “no” points at something reproducible rather than just a rating.