What is Deep Research?

AI deep research is an agentic workflow that plans your question, searches across dozens of sources, extracts and cross-checks findings over many rounds, and composes a cited report. It's the difference between getting links and getting a finished brief.

In plain English

Deep research is the difference between getting a list of books and getting a well-read friend who summarizes them, cites the good parts, and tells you what's still unresolved. Both are useful. Only one saves your evening.

Googling a hard question is asking the library to point at shelves. Deep research is the library writing your lit review.

The problem a single search can't solve

A regular web search returns ten ranked links. Useful — but links don't synthesize. To answer a genuinely hard question — "what's the state of the art in battery recycling for EVs?" — you'd open dozens of pages, reconcile conflicting numbers, track which sub-questions are still open, and write it all up.

That's hours of mechanical work. Deep research automates exactly that mechanical layer.

How a deep research agent actually works

Deep research runs a loop, not a single query. In Orbit, the research engine follows an iterate-research pattern:

• Plan — the agent breaks your question into sub-questions.
• Classify and generate queries — it produces targeted search queries for each gap.
• Search — it queries multiple engines, including premium answer providers.
• Extract — it pulls content from promising pages.
• Merge — findings are delta-merged into a growing research draft.
• Coverage check — it marks questions answered and re-targets the rest.
• Repeat — typically 8–14 rounds across 65+ sources.

Only after the rounds converge does the agent compose the final report — plus a chat brief you can read immediately.

Depth levels matter

Not every question needs maximum depth. Orbit offers two modes:

• Fast — a light pass that answers straightforward questions in a few minutes.
• Deep — the full engine: 8–14 rounds, 65+ sources, and a long-form report with source annotations.

Choosing depth is a cost trade-off. Deep runs burn more tokens across more models, so the interface lets you pick before you commit.

What the output looks like

A finished deep research run gives you more than a wall of text:

• A final report written to the project as a durable artifact.
• An HTML report with the sources' images and links.
• A coverage checklist showing which sub-questions were answered.
• Per-source provenance so you can verify claims yourself.
• A Dig deeper continuation that re-runs the engine on your follow-up.

And because research lives inside a workspace, the findings feed directly into whatever you build next.

A note on reasoning power

Research quality depends on the reasoning model doing the planning and writing. Orbit routes across 150+ models, and its deep research mode uses strong reasoning models for planning, query generation, and the final report while keeping cost-sensitive steps on faster models.

When deep research is worth it

Use it for market research, technical landscape reviews, competitor analysis, academic groundwork, and any question where a single answer is suspicious. Skip it for "what's the capital of X" — that's what a search is for. The cost of deep research only pays off when synthesis is the hard part.

Try it and compare

Take a real work question — a market, a competitor, a technical landscape. Run it through a web search, a chatbot, and Orbit's deep mode. Compare how many sources each cites, which sub-questions each answers, and which output you could actually forward to a colleague without embarrassment.

FAQ

Q: What is AI deep research? A: AI deep research is an agentic workflow that plans a research question, runs many targeted searches across dozens of sources, extracts and cross-checks findings, and composes a cited report.

Q: How is deep research different from a web search? A: A search returns links ranked by relevance. Deep research iterates: it generates queries, reads and extracts content, merges findings across rounds, and tracks which questions are answered and which are not.

Q: How long does deep research take? A: It depends on depth. A fast pass can finish in a few minutes, while a deep run in Orbit spans 8–14 rounds across 65+ sources and can take well over an hour before the report is written.

Q: Can deep research replace doing your own research? A: It replaces the mechanical parts — finding, extracting, and summarizing sources. You still set the question, judge the evidence, and decide what matters.

Build it on Orbit

Run true deep research — 8–14 rounds across 65+ sources — inside the same workspace where you chat and build apps. Start building on Orbit.