Research Mode is in public beta on Plus and Pro plans.
Most AI chat surfaces hit a wall at the report length most knowledge work actually needs. You ask a deep question; you get a competent two-paragraph summary; you wish for the longer thing.
Research Mode is the longer thing. It produces 60+ page reports with inline citations sourced from real web results, organized into chapters, with executive summaries and references sections that link back to the source material.
How it works
Research Mode is a hybrid of two ideas from the published literature:
- STORM (Stanford) — multi-perspective question generation. The system spawns "personas" that ask the same topic from different angles, then synthesizes their findings into an outline.
- IterResearch — iterative refinement. Drafts are critiqued, gaps are identified, the loop runs again.
The two are stitched together by the same ai_proxy tool loop that powers the rest of vMira chat, so the citations and tool calls behave the same way the user already knows.
Trade-offs
- Reports take 10–25 minutes to generate, not seconds.
- Plus and Pro plans only — the compute cost is real.
- Output is read-only by default; an "edit and re-run a chapter" workflow is on the roadmap.
Open the Research tab in a chat to try it.