Deep Research
For complex questions that require evidence from multiple documents, Deep Research performs iterative retrieval, automatically identifying gaps in the initial evidence and searching for additional sources to build a comprehensive answer.
When to use Deep Research
Deep Research is most valuable when:
- Your question spans multiple papers (e.g., “How do different studies measure cognitive decline in aging populations?”)
- You need a comprehensive literature synthesis
- You want to identify contradictions or consensus across your library
- The initial answer feels incomplete or missing important perspectives
For single-document questions or factual lookups, standard chat is faster and more efficient.
How it works
Initial evidence retrieval
Virza searches your workspace for passages relevant to your question, just like standard chat.
Gap analysis
The AI analyzes the initial evidence and identifies what’s missing:
- Are there methodological perspectives not yet represented?
- Are there contradicting findings that should be surfaced?
- Are key statistical claims missing supporting evidence?
Follow-up retrieval (up to 3 rounds)
For each identified gap, Virza generates targeted follow-up queries and retrieves additional evidence from your library. Each round adds new sources that address specific gaps.
Synthesized response
The final answer integrates evidence from all retrieval rounds, with citations linking to specific source passages.
The reasoning chain
When Deep Research is active, a collapsible timeline shows each step of the process:
- Round 1: Initial query → X sources found → gap identified: “Missing methodology comparison”
- Round 2: Follow-up query → Y additional sources → gap identified: “No contradicting evidence”
- Round 3: Final query → Z sources → evidence sufficient
Each entry shows what gap was found, what query was generated, and what was retrieved. This gives you full auditability of the reasoning process.
Enabling Deep Research
- Open any chat conversation
- Click the Deep Research toggle in the chat toolbar (or the beaker icon)
- Ask your question
- Watch the reasoning chain expand as evidence is gathered
Deep Research uses more AI credits than standard chat because it performs multiple retrieval and analysis rounds. A typical deep research query uses 3–5× the credits of a standard chat message.
Tips for effective deep research
- Ask broad, analytical questions: “What are the competing theories on X?” works better than “What does Smith 2024 say?”
- Use workspace scope for cross-document synthesis
- Use collection scope to limit deep research to a specific research area
- Review the reasoning chain: it shows you exactly how the AI built its answer, and you can verify each step
- Follow up: Deep Research remembers the conversation context, so you can ask narrowing questions after the initial synthesis
Limitations
- Deep Research performs up to 3 retrieval rounds. Very complex topics may still have gaps after 3 rounds.
- Quality depends on your library’s coverage. If your library doesn’t contain relevant papers, Deep Research can’t find them. Consider using the Discover tab to import additional sources.
- Each round adds latency. A deep research response typically takes 10–30 seconds to fully complete.