Deep research is an AI workflow where an agent investigates a question over many sources instead of answering from a single model response. In Scholarly, you ask a question — optionally grounding it in your own documents — and the agent plans the investigation, searches and reads in the background, and returns a report with inline citations, sized as a one-page brief, a decision memo, or a full investigation.
The difference between deep research and a good chat answer is process. A chat reply is one pass over what the model already knows plus a quick search. A deep research run starts by pinning down what you actually need — the agent asks a couple of clarifying questions when the goal is ambiguous — then works through sources methodically, comparing what they say, noting where they disagree, and assembling the answer with the evidence attached.
That process matters most for questions where being roughly right isn't enough: what the trials actually showed, which vendor survives a real comparison, what a regulation requires, where the market has moved since last year. Those are questions you'd otherwise answer with an afternoon of open tabs and a document of pasted quotes.
Most deep research tools only search the public web. Scholarly's runs inside your workspace, so an investigation can start from the material that actually defines your question: the internal report that needs external context, the vendor proposals you're comparing, the draft that needs its claims checked. Attach up to ten sources — uploaded files, workspace folders, saved pages, even earlier research sessions — and the agent treats them as primary evidence.
The evidence filters steer everything beyond your documents. Reviewing clinical or academic questions, favor papers and journals. Working on operational or technical material, favor handbooks and manuals. Tracking a market, favor news and industry sources. Checking compliance, favor government and policy publications.
Use chat when you need an answer now and one good pass is enough — a definition, a quick explanation, a summary of a document you've linked. Chat is immediate and conversational, and for most questions it's the right tool.
Use Deep Research when the question deserves evidence: evaluating options before a decision, reviewing what's actually known about a topic, checking a claim you're about to repeat, or building the cited document a colleague or client will read. The output is an artifact — a report you can share and return to — rather than a reply that scrolls away.
A finished investigation stays live in your workspace. Chat with it to question findings or pull out specifics, queue follow-up reports in the same session as the question sharpens, and hand the results onward — turn the report into slides for a briefing, a podcast for the commute, or study materials for a team that needs to learn it.
