Home/Docs/Research Mode
🔍Deep AI research

Research Mode

Research Mode turns one question into a comprehensive, cited report. The AI plans a strategy, searches the web in parallel, scores each source for credibility and relevance, and synthesizes the findings — with numbered citations and a visible thought tree of every step it took.

What is Research Mode?

Research Mode goes well beyond a single web search. Instead of returning a list of links, the AI runs a multi-stage investigation: it plans sub-queries, searches in parallel, critically evaluates each source, and synthesizes a structured report with inline citations. It can also search your organization's own documents alongside the web, so proprietary context informs the conclusions.

Open Research from the Knowledge, or launch it as a side panel from any AI Chat session via the Deep Research tool.

ℹ️
Powered by Brave Search

Web search in Research Mode uses the Brave Search API. Research runs are capped by a monthly per-user limit — 50 by default — rather than the per-organization spend budget that governs web search elsewhere in YOffice.

The research process

Research follows a four-stage loop. Progress is shown live, and each step is recorded in a thought tree you can expand.

🧭PlanBreak into sub-queries
🌐Web searchBrave Search
🔬Review findingsCheck gaps & conflicts
🧩Synthesized reportWith citations
One question becomes a planned, sourced, cited report
  1. Planthe AI decomposes your question into sub-queries to cover different angles and source types.
  2. Searchsub-queries run one at a time via Brave Search; the AI prioritises authoritative, recent, relevant sources.
  3. Evaluatethe AI reviews the combined results in one pass for gaps, conflicts, and credibility before writing the report.
  4. Synthesizeaccepted findings are merged into a structured report with numbered citations, contradictions flagged.
Thought tree
🧭 Plan: 3 sub-queries
Hardware advances
Error correction
Industry use cases
🌐 Search (18 sources)
🔬 Review findings
🧩 Synthesize
Quantum computing in 2026Research
Summary

Below-threshold error correction was demonstrated by multiple labs; broad commercial use remains a few years out, while optimization and simulation use cases advance fastest.

Findings
1.Logical-qubit error rates dropped below threshold [1]
2.1000+ qubit systems shipped with better coherence [2]
3.Pharma begins quantum simulation for discovery [3]
Sources: [1] nature.com · [2] research.ibm.com · [3] mckinsey.com · +15
Research Mode in progress — thought tree of steps (left), live status and the cited report (right)

Thought-tree visualization

Research isn't a black box. As the AI works, it builds a thought tree — a structured view of its plan, sub-queries, the sources it pulled, and the reasoning at each step. Expand any branch to see exactly how a conclusion was reached and which sources supported it.

Report structure

SectionDescription
SummaryA short overview of the key conclusion for a non-specialist reader.
FindingsNumbered insights, each with an inline citation reference.
ConfidenceA percentage showing how strongly the sources agree — a useful flag for findings worth a closer look.
SourcesCited sources with titles, domains, and URLs, numbered to match.

Continuation and history

Research tasks are saved so you can come back to them. From the history you can re-open a past report, and you can continue a prior task — the AI picks up with the earlier findings as context and digs deeper or branches in a new direction.

ℹ️
Research has a monthly limit

To keep costs predictable, both starting a new research task and continuing a past one draw on the same monthly per-user limit — there's no separate allowance for continuing.

Search the web + your Knowledge Base

Research Mode can search your organization's own documents alongside the web — useful for competitive analysis or any topic where your team's existing knowledge should inform the answer. Link a knowledge folder to your chat session before starting and the AI will:

  • Automatically search the linked folders with semantic retrieval (documents, transcripts, Canvases).
  • Search the web with Brave for external context and recent developments.
  • Merge both into one cited report.
💡
Combine with meeting transcripts

Link your transcripts folder to the session and ask, "What are the themes from our last three discovery calls, and how do they compare to current industry trends?" The AI cross-references your transcripts with fresh web research in one report.

Exporting reports

  1. Open the report

    Open a completed research task from the chat or the Research history.

  2. Export to PDF

    Download the full report as a formatted PDF to share outside YOffice.

  3. Reuse it

    Copy the findings, save them to a Knowledge folder, share them to a chat, or add them to Memory — or keep the saved task for later continuation.

ℹ️
Good to know

Research draws on live web search, so it needs an internet connection even when paired with a self-hosted model. Results reflect publicly accessible content and the freshness of what's online, and any low-agreement findings make great starting points for a closer look.