Ask a hard question, get a research report
Deep Research sits next to web search in the Pervagans composer. Switch it on and the turn stops behaving like a chat reply: it runs several searches across different angles, reads across them, and writes a structured report with an executive summary, ordered sections and a conclusion. Every factual claim carries a link to the source it came from.
3 credits a run · Signed-in feature · Free plan includes 200 monthly credits
A report with a shape you can rely on
A chatbot answer is whatever length the conversation drifted toward. A Deep Research report follows the same four parts every time, so you can skim it, hand it to someone else, or paste it into a document without reformatting it first.
A title
The report opens on its own heading. No preamble, no restating of your question, no description of what the model is about to do.
An executive summary
Three to six sentences with the bottom line up front, so the answer is readable before you decide whether to read the rest.
Ordered body sections
Short, descriptive headings in a logical order, with a Markdown comparison table wherever options, tools, vendors or metrics are being weighed against each other.
A conclusion
A closing section that lands the argument, rather than a sign-off or a promise to help further.
One callout, at most
A single highlighted callout is allowed, and only for a genuine risk or caveat you should not miss. Nothing is highlighted for decoration.
A diagram when it earns its place
On models that support it, the report may include one Mermaid diagram where a process or a relationship is genuinely clearer as a picture than as prose.
Research behaviour, not a longer answer
The difference is in what happens before the first word is written, and in what the finished document is willing to admit.
It searches several angles before it writes
A quick answer summarises the first page it finds. Deep Research runs several distinct searches covering different sub-questions, different viewpoints and recent dates, then synthesizes across everything it gathered instead of paraphrasing sources one at a time.
- Several distinct searches per run, not one lookup
- Different sub-questions and opposing viewpoints covered
- Concrete numbers, dates and prices wherever they exist
- Synthesis across sources, not a chain of summaries
Every claim carries its source, right there
Each factual claim is followed by a link whose visible text is the source's bare domain, so you can see who said it without clicking anything. When several sources back one fact, their links sit side by side.
- Citations sit immediately after the claim they support
- Link text is the source domain, so provenance is readable
- Each citation points at the specific page, not a homepage
- No trailing sources list, because nothing was deferred to it
It says where the sources disagree
Real evidence rarely lines up. The report compares and contrasts the sources it found, names the ones that conflict, and states plainly where the evidence is thin instead of smoothing the gap into a confident sentence.
- Conflicting sources are set against each other, with numbers
- Uncertainty is stated rather than hedged into vagueness
- Gaps in the evidence are named as gaps
- No filler, no boilerplate caveats, no restating the question
A document to read, in the browser or on your desktop
Deep Research never returns an interactive panel or a mode-switching widget. It returns headings, prose, tables and links, so the report reads the same in a browser tab and in the native Windows app, and copies cleanly into wherever your work actually lives.
- Markdown headings, tables, links and at most one callout
- Identical in the browser and in the Windows desktop app
- Save it as an Artifact to keep a versioned copy
- Keep it inside a Project so later questions build on it
Priced higher, and worth knowing up front
A Deep Research run does several searches and writes a long document, so it is not billed like a one-line reply. Here is the whole story before you turn it on.
3 credits, against 1
Each Deep Research turn draws 3 credits from your monthly pool. An ordinary chat turn draws 1, and so does a chat turn with plain web search on.
Signed in only
Deep Research is off for anonymous visitors. Creating a free account takes no card, and the Free plan comes with 200 credits a month to try it with.
Web search comes with it
Switching Deep Research on switches web search on too. You do not need to set both, and the report is grounded in live results either way.
Credits are shared across everything you make: chat, research, images, decks. Free is $0 for 200 credits a month, Pro is $20 for 1,000 and Max is $167 for 8,350. Compare them on the pricing page.
Research is one of four things you can make
Deep Research shares an account, a credit pool and a set of models with the rest of Pervagans, so a report can turn into a deck, or become the grounding for a Project, without leaving the tab.
Presentation
Describe an idea and get a complete editable deck, real slides with real text rather than a picture of a slide. 17 layout families, 110 themes, and exports to PPTX, PDF and Markdown.
Deep Research
The structured, inline-cited report this page is about. One toggle in the composer, 3 credits a run, available on every text model your plan unlocks.
Projects
Per-project files, custom instructions and durable memory. Upload your documents and every answer in that project is grounded in them, up to 100 MB per file and 50 files per project.
Artifacts
HTML, SVG, documents and code the model writes for you, rendered live and saved with an append-only version history plus an optional public share link.
All four run in the browser and in the native Windows desktop app, which loads pervagans.com live so it is current the moment the product changes. Reports come back in the language you asked in, Arabic included, with the summary and conclusion named natively (ملخص تنفيذي and الخلاصة) and the whole document laid out right to left.
Questions, answered
What Deep Research does, what it costs, and what it will not pretend to know.
What is Deep Research in Pervagans?
Deep Research is a toggle in the Pervagans composer, next to web search. With it on, a turn stops behaving like a chat reply and returns a written report instead: a title, an executive summary of 3 to 6 sentences that leads with the bottom line, body sections in a logical order, and a conclusion. Every factual claim carries an inline citation to the source it came from.
How is it different from turning on web search?
Web search grounds an ordinary chat answer in live results. Deep Research changes the shape of the work: it runs several distinct searches covering different sub-questions and viewpoints before it starts writing, synthesizes across those sources instead of summarizing one page, says where the sources disagree, and returns a structured report rather than a conversational reply. Turning Deep Research on turns web search on with it.
How much does a Deep Research run cost?
A Deep Research turn costs 3 credits, against 1 credit for an ordinary chat turn. Credits come from the same monthly pool as everything else in Pervagans: 200 credits a month on the Free plan at $0, 1,000 on Pro at $20 a month, and 8,350 on Max at $167 a month.
Do I need an account?
Yes. Deep Research is switched off for anonymous visitors, so you need to be signed in for the toggle to take effect. The Free plan is enough to try it and takes no credit card.
Does the report end with a list of sources?
No, and that is deliberate. Citations are inline, placed immediately after the claim they support, written as a link whose visible text is the source’s bare domain. You check a fact where you are reading it rather than counting footnotes at the bottom, and there is no trailing sources list to scroll to.
Which models can I run it with?
Any of the 10 text models your plan unlocks, including Claude, GPT, Gemini, Grok and Qwen. The report contract is the same whichever one answers: same structure, same inline citations, same tables where a comparison calls for them.
Can I get a report in Arabic?
Yes. Reports come back in the language you asked in, Arabic included, with the sections named natively in Arabic and correct right-to-left layout for the whole document.
Does it work in the desktop app?
Yes. Pervagans runs in the browser and as a native Windows desktop app, and Deep Research behaves identically in both. The desktop app loads pervagans.com live, so a product change reaches it without an update step.
Bring it your hardest question
One toggle, several searches, and a report that shows its sources in the sentence rather than in a footer. Try it on the Free plan, on the web or in the Windows app.