Evidence sits where you write
The material you gathered is one sidebar tab away from the manuscript that depends on it. Checking a figure mid-paragraph does not mean leaving the workspace or hunting through a downloads folder.
Import PDFs, Word files, and spreadsheets, transcribe scanned pages when you choose, and ask questions answered with page citations you can check.
Start writing free
ROUGH NOTES → STRUCTURE → LONG FORM
Imported files become project sources inside a project. Text is extracted the moment you upload, page boundaries are preserved for PDFs, and every PDF gets a reader where the original sits next to the text pulled out of it, page by page, each one labelled with its number.
Research is only useful if you can get back to the exact page, which is the discipline behind citation management. A folder of downloads cannot tell you which report held the figure you half remember, and a summary you cannot trace is not evidence. Sources here keep their pagination, so a claim can always be checked.
Upload as many as ten files at once in TXT, Markdown, CSV, PDF, DOCX or XLSX. Spreadsheets are flattened sheet by sheet and CSV rows become readable lines, so tabular research is searchable rather than opaque.
The Source Reader shows the original PDF beside the extracted text, split page by page. A page that turns out to be a scan is marked as waiting rather than quietly returned empty, so nothing goes missing unnoticed.
OCR is opt-in and never runs on upload. When you approve it, Strut AI transcribes the blank pages in reading order, up to sixty pages in one run, reserving 2,000 tokens per page. Pages with selectable text are skipped.
Ask a source a specific question and the answer is drawn from that file alone, with a page citation behind each factual statement. Where the source does not support an answer, it says so instead of filling the gap.
Save a passage as a rough note carrying the file name and page, and let the project's sources join your writing operations as context, up to eight files for each request, so evidence follows the prose.
The material you gathered is one sidebar tab away from the manuscript that depends on it. Checking a figure mid-paragraph does not mean leaving the workspace or hunting through a downloads folder.
Page citations mean an assertion is checkable in seconds against the file you uploaded. That is the difference between research you can publish from and research you can only browse.
Nothing is transcribed behind your back. The reader states how many scanned pages are waiting and what they will reserve, then waits for you to press the button before a single page is processed.
A PDF with no text layer imports and is flagged as containing scanned pages, with a count of how many. The selectable parts of the same file stay usable straight away, and you decide separately whether to transcribe the rest.
The question is answered from that one file, and each factual statement must carry a page reference. Any instruction hidden inside the document is ignored rather than acted on, so a source cannot redirect the answer.
A single upload request is capped at 25 MB in total, which comfortably covers a batch of reports or a long manuscript. Split a very large scan into parts if it goes beyond that ceiling.
Each file belongs to the project it was imported into, which keeps unrelated research out of your AI context. Workspace search still reaches across every project you own or collaborate on, sources included, so nothing is stranded.
Give your rough notes
somewhere to go.
Projects, notes, versioned drafts, imported sources, and careful AI assistance in one private workspace.