Deep Research Hub

The Deep Research Hub runs a full end-to-end research pipeline on any topic you provide. It generates a research plan, searches across multiple academic databases, filters and clusters results by theme, and synthesizes a coherent summary — all with a live progress panel you can follow in real time.

How it works

The pipeline runs in six stages:

1. Query — You enter a research question or topic.
2. Plan — The AI generates a structured research plan with sub-questions and search strategies. You review and approve (or edit) before the search begins.
3. Search — The agent queries all enabled academic sources in parallel.
4. Filter — Results are scored for relevance and duplicates are removed.
5. Cluster — Papers are grouped into thematic clusters automatically.
6. Synthesize — The agent writes a structured summary grounded in the retrieved papers.

Plan-first workflow

Before any search begins, the Deep Research Hub shows you the research plan it intends to follow. The plan includes:

• The main research question broken into sub-questions
• Planned search queries for each sub-question
• Which databases will be queried
• Estimated scope

Review the plan and edit it if needed, then click Start Research to launch the pipeline. This lets you steer the direction before committing to a potentially long search.

Adaptive research agent

The research agent is LLM-driven and adapts as it works. After each search round it:

• Detects gaps in coverage based on what was found
• Expands queries to cover under-explored angles
• Decides whether additional search rounds are needed
• Stops when coverage is sufficient or when the session limit is reached

You do not need to supervise the agent during a session. Return when it finishes and review the progress panel for a full log of what was searched and found.

Academic sources

The Deep Research Hub searches six academic databases simultaneously:

Results from all sources are deduplicated and merged into a single result set before filtering.

Research progress panel

While a session runs, the progress panel shows live status updates:

• Which stage is currently active
• Papers found per source per round
• Gaps detected and new queries generated
• Cluster labels as they form
• Synthesis progress

You can collapse the panel and return to other work. The session continues in the background.

Research sessions

Each research run is saved as a session. Sessions are:

• Resumable — If a session is interrupted, open it again to continue where it left off.
• Cloud-synced — Sessions are available on all your devices.
• Persistent — Results, clusters, and synthesis are saved indefinitely.

Open a past session from the Research sidebar to review its results, re-run synthesis with updated settings, or generate a literature review from its paper set.

Tier limits

A session counts when you approve the plan and start a search run.

Starting a session

1. Open the Deep Research Hub from the sidebar or from the command palette (Cmd+P (Windows: Ctrl+P)).
2. Enter your research question in the query field.
3. Review the generated plan. Edit any sub-questions or queries if needed.
4. Click Start Research.
5. Watch the progress panel or return later when the session completes.

Tip: Narrow your research question before starting. Broad questions like "machine learning" produce many loosely related papers. Specific questions like "transformer architectures for time-series forecasting" produce tighter, more useful clusters.

Related

• Academic Paper Search
• Literature Review Generation
• Knowledge Graph Viewer
• Bibliography System