What deep research does that a normal answer does not
A normal chat reply answers from what the model already knows, which can be months out of date and comes with no way to check it. Deep research goes and looks. It turns your question into a handful of search queries written from different angles, picks the most useful results, reads each page rather than its snippet, and takes notes that stay attached to the page they came from.
After the first pass it checks what is still missing and runs at most one more round of searches to fill the gaps. Then it writes a brief from its notes. The reference list is built from the pages it read, not from the model’s memory, so a citation cannot point to a page that does not exist. You watch each step while it happens: the searches it planned, the pages it opened and the citations it checked.
It is useful when the answer is spread across many sources: comparing products or services, catching up on a news story, checking what the evidence says about a health or money question, or getting a first map of a subject before you read further.
Run deep research in three steps
Step 1: Open nokeep in Chat mode, press the + next to the message box and choose Deep research. Without a plan or credits the option shows a lock, and tapping it opens the plans.
Step 2: Ask one clear question and name the subject. “Compare the three cheapest ways to send money from the UK to India, with fees and speed” works far better than “research this”. If you care about a country, a year or a budget, say so.
Step 3: Send, and let it work. A run is slower than a normal reply because it reads ten pages rather than one. When the brief arrives, open the sources that matter most and ask follow-up questions in the same chat.
What it costs, and which plan you need
Deep research is not part of the free tier, because one run is fifteen model calls plus the searches and page reads behind them. With credits and no plan, a run costs about 150 credits, about 15 cents, where 1,000 credits are one dollar. The composer shows the price before you send, and the run settles to what it actually used. Top-ups start at $5, enough for about 33 runs, and bought credits never expire.
On a plan (Plus, Pro or Max) deep research is included and does not use your credits. Plain web search, which gives a quick answer with sources inside a normal chat, comes with a plan or credits and never uses credits: with credits but no plan, both work.
Buying credits or a plan gives you an account key instead of a login. There is no email to confirm and no password. Save the key, because it holds your balance.
What stays private
Your question is not pasted into a search engine. It is rewritten into short, neutral search queries, so a search provider sees “first trimester medication safety data” rather than the first-person sentence you typed, and the websites that are read never see you. The pages it gathers live only for the length of the run and are thrown away with the request.
The finished brief is saved in your browser like any other chat. nokeep keeps no copy of your question or the brief on its servers and does not train on them. Because the chat lives in your browser, clearing site data removes it, so export or back up a brief you want to keep.
Limits
A run reads at most 10 pages, takes at most 2 from any one site so a single source cannot dominate, and searches in at most two passes. That keeps it fast and affordable, but it is not a systematic review. For a decision that matters, read the key sources yourself.
It can only read pages that are open to the public. Paywalled articles and pages that need a login are out of reach. If nothing useful turns up, it says so rather than inventing an answer, and you can try naming the subject more precisely.
A brief is only as current and as accurate as the pages it found. Citations show where a claim came from; they do not prove the page is right.