There is a large gap between “search the web” and “research this properly.”
A quick search can find a current fact. It is not enough for a market landscape, a policy brief, a technical comparison, or a question where the sources disagree. Those jobs need a plan, several search directions, actual page reading, and a result you can inspect later.
That is what Deep Research in Aii Cloud is for.
Deliberate by design
Deep Research is never triggered by a keyword and never spends a research budget because you pasted a URL. You arm it explicitly for one message from the composer—or type /deep-research at the start of a draft.
The selection applies to that send only. Aii switches the turn into Work, checks that the chosen model and balance can support the run, then shows the research plan before going deep.
This one-turn design is intentional. “Research this” is a spending and time decision, not a hidden mode the product should guess on your behalf.
Search, read, and keep receipts
A Deep Research run does more than collect snippets. It can:
- break the question into research threads;
- search in waves as the evidence changes the plan;
- fetch and read the relevant pages;
- compare claims across sources; and
- write a single report with numbered citations.
Progress stays visible while the run works: the plan, current phase, searches used, and sources collected. If the available budget is reached, Aii writes from the evidence it already gathered instead of ending with a dead stream.
Every report arrives as a versioned Markdown artifact in the workspace. Citation markers resolve against the adjacent Sources ledger, so you can move from a conclusion to the page behind it. Invalid citation numbers are removed rather than presented as evidence.
If only a few useful sources exist, the report says so and stays short. Thin evidence does not get padded into false confidence.
Inspect a real research run: Read Aii's NASA and NOAA aurora comparison. The read-only page includes the cited report, a downloadable Markdown file, and a Sources ledger you can open to check the underlying pages.
Explore the result as a website: Open Aurora Explorer, a visual exhibit Aii built from the research. Select a step to see how an aurora forms, compare NASA's and NOAA's explanations, and follow the links to their original sources. The illustration is static and is not a live forecast.
A report you can keep working with
The research result is not a disposable answer bubble. It has the same Preview, Code, Diff, download, and version history as other Aii Cloud artifacts.
Follow-up turns are ordinary Work messages:
“Condense section three for an executive audience.”
“Turn the risks into a table.”
“Challenge the strongest assumption in the recommendation.”
Those edits do not silently launch another research run. If you want a fresh evidence sweep, arm Deep Research again. That creates a new research budget, a new report version, and a new citation space.
Honest limits
Research can take time, and the browser must stay connected while it is running. Aii tells you that before the run starts.
Stopping or losing the connection preserves the sources collected so far and records the run as interrupted. Search-provider trouble produces a reduced-sourcing note. A model that is not strong enough for the research lane is refused before charging, with a prompt to choose an eligible one.
Try it with a question that has edges
Open Aii Cloud, choose Deep research, and ask something where recency, disagreement, or primary evidence matters:
Compare the strongest arguments for and against adopting passkeys first
for a small B2B SaaS. Separate standards evidence, implementation risk,
and assumptions. End with a staged recommendation.A good research product should make it easier to challenge the answer, not merely easier to generate one. The report is the result; the source trail is the reason to trust it.
Deep Research is a deliberately armed, credit-metered Work capability in Aii Cloud.