Deep Search is a multi-step research mode for Agent chat. Instead of answering in a single pass, it plans its approach, works through your documents and sources in multiple rounds, and synthesizes a thorough, well-grounded answer. It's the right tool for complex questions that span many files or require working through large documents. Here's how to turn it on and when it's worth using.
What is Deep Search?
Deep Search runs multiple passes across your content, aggregating and analyzing as it goes. It reads source material in depth rather than skimming, reasons across documents, and produces a more comprehensive answer than a single-pass response.
Compared to a standard Agent chat response, Deep Search:
Uses more of the relevant sources in your library, not just the top matches
Reasons more deeply across documents before answering
Produces longer, more structured responses with broader citation coverage
Takes longer to run than a standard response
Tip: Deep Search shines on research-style questions: literature reviews, multi-document synthesis, chronological summaries of large document sets, and any question where you'd rather wait a bit for a complete answer than get a fast partial one.
How to enable Deep Search
Deep Search is a toggle in the chat input. To turn it on:
Open Agent chat.
Click the + (plus) button on the left of the chat input.
Find Deep search in the menu and flip the switch on.
Once enabled, Deep Search stays on for the rest of that conversation. Your next message in that chat will use Deep Search automatically. To turn it off, open the plus menu again and flip the switch off.
Note: Deep Search is available on the Max plan. If your plan doesn't include it, you'll see a Max tag next to the toggle. Upgrade from Settings > Billing to unlock it.
When to use Deep Search
Use Deep Search for:
Literature reviews across many papers in a folder
Comparing findings, methods, or results across multiple studies
Summarizing a large document set into a single coherent narrative (for example, a patient record, a case file, or a book)
Research questions where completeness matters more than speed
Questions that need evidence from across your library, not just the most obvious sources
Skip Deep Search for:
Quick factual lookups
Simple chats where a standard response is enough
Creating notes, citations, or images (those tools don't need Deep Search to work)
Follow-up questions on a conversation that's already covered the ground
Tip: You can start a chat with Deep Search off, ask an exploratory question, then toggle it on when you want to go deeper on a specific thread. It applies to your next message, not past ones.
Scoping Deep Search
Deep Search respects the same scope controls as the rest of Agent chat. You can point it at just your library, just the web, a specific folder, or a specific file.
To narrow the scope:
Open the file or folder you want the search grounded in. Agent chat is aware of what you have open.
Use the library search and web search toggle buttons below the chat input to include or exclude each source.
Type @ to @mention specific files, folders, or connectors you want included.
Tip: For literature reviews, open the folder containing the relevant papers before turning on Deep Search. It will use that folder as its primary context and work through the documents systematically.
Trade-offs to know
Deep Search trades speed and cost for depth:
Slower responses. A Deep Search answer can take noticeably longer than a standard response. You'll see progress as the AI works through your documents.
Longer outputs. Answers tend to be more detailed and structured. If you want a shorter response, ask for one explicitly in your prompt.
Reach for Deep Search when you need thorough, multi-source synthesis. For quick factual lookups or simple follow-ups, the standard Agent chat response is faster and does the job.