Ask about a competitor’s pricing, a library’s breaking changes, or what shipped in a spec last month, and the AI searches the live web instead of guessing from training data. The answer comes back in the thread, next to everything else you already know.
An assistant that only knows your workspace stops at the workspace boundary, and one that only knows its training data confidently tells you about a version that shipped two years ago. Both send you to a browser tab, where the answer arrives without any of your context around it.
Pricing, APIs and competitors all changed since the model was trained. Nothing in the answer says so.
You research in a browser, decide in your head, and the reasoning lands nowhere the team can read it.
Six tabs of changelog to answer one question about whether it’s safe to upgrade.
In chat, in a comment, in a page composer, or by ⌘O. No mode to switch into.
The workspace comes first; the web is for what the workspace doesn’t hold. You see the research step happen in the activity pill.
In the thread, alongside your decisions and tasks — so “given what we settled in D-8, is this still a problem?” is one question, not two.
“Scan our competitors’ changelogs and lead with what changed” works as a weekly schedule, posting into chat or writing a page.
Ask for a page and the findings become a real doc — with tables and charts — instead of a message that scrolls away.
Whether an agent may search the web is one of its settings, and an admin can switch the capability off entirely in Agents → Tools.
Web research is a built-in tool, not an add-on. It spends AI credits like any other AI work.
Free for up to 10 people, with 1,000 AI credits included. No credit card, no sales call, no onboarding deck.