Research a topic and see where the answer came from
Turn on WebSearch, ask the question, and get an answer built from pages the model actually read, with the URLs to check.
Try this promptA model on its own answers from training data that stopped months ago, and it will not always tell you that. For anything current, specific, or worth citing, that is not good enough. You want the model to go and look.
On Berges AI that is the WebSearch pipeline. Pick it from the sidebar and the model gets two tools: search the web, and fetch a page to read it in full. Every search and every fetch shows up as a badge on the reply, so you can see what it used.
Same task, less slop
Great question! Lithium battery rules can vary, so it's always best to check with your airline. Generally speaking, lithium batteries are subject to restrictions and may not be allowed in checked luggage. Be sure to consult the relevant regulations before you travel. Safe travels!
Spare batteries and power banks: carry-on only, never checked. Devices with a battery installed can go in checked bags if switched off. Up to 100 Wh needs no approval; 100 to 160 Wh needs airline approval, two spares max; above 160 Wh is not allowed. Sources: EASA passenger guidance on lithium batteries, IATA dangerous goods page for passengers.
Research with sources with Berges AI
Select WebSearch
Make sure WebSearch is the pipeline selected in the sidebar. Nothing you type reaches a search provider until the model decides a question needs a search, and each search shows as a badge on the reply.
Ask for sources
Say what you want to know and add "search for primary sources and cite them". The model searches, picks the promising results, fetches them, and reads the actual text rather than the snippet.
Check the badges, then the links
Each reply shows what it searched and what it fetched. Open the URLs before you rely on anything. The model can still misread a page. The citations are there so you do not have to take its word for it.
Pair it with the Berges WebSearch pipeline
WebSearch gives the model a search tool and a page-fetch tool and lets it decide when to use them. The model doing the research is a reasoning model, so it weighs what it finds before it answers.
Put this in your prompt
- → The exact thing you need to know, with the dates, versions, or jurisdictions that narrow it down.
- → Whether you want official sources, recent news, or the original paper. Say which.
- → A URL, if you already have one. Paste it and the model reads that page directly.
- → What "current" means to you: this week, this year, since a particular event.
Things people ask
How is this different from asking without WebSearch?
Without WebSearch the model answers from memory, which stops at its training cutoff and can be wrong about specifics. With WebSearch it reads live pages and tells you which ones. Use WebSearch when the answer depends on today's state of the world or when you need to cite it.
Who sees my searches?
When WebSearch is on, your search queries go to a third-party search provider (currently Brave Search) to return results. Fetched pages are read for that turn and not stored. Your messages and the reply are stored only if you save the conversation, encrypted at rest like everything else.
Can it read a specific page I give it?
Yes. Paste the URL in your message and the model fetches the page text, on any pipeline, not just WebSearch. It works for public pages up to a size limit; paywalled or login-only pages will not load.
Will it be slower?
Yes, by a few seconds per search or fetch. That is the cost of reading rather than guessing. For a question that does not need fresh information, pick another pipeline and get the faster answer.