Turn a folder of papers into one context window.
Drop the PDFs, your notes, and the data. FileConcat pulls the text out of every paper right in your browser, packs it into one document, and shows you the token count so you know it fits ChatGPT, Claude, or Gemini before you paste.
- Nothing is uploaded
- Token count shown before you paste
- PDFs, notes, and data
Drag your reading folder here
Papers, notes, and data. Read in a second.
Know it fits before you paste.
The hard part of reading with a model is not the reading, it is the budget. FileConcat counts the tokens as it bundles, so you see whether the whole pile lands in one window before you ever paste it.
Counted locally, with the same tokenizer the models use
From a reading pile to a prompt.
- 1
Drop the reading folder
Papers, notes, and datasets together. Nested folders are fine.
- 2
Every file becomes text
PDFs are extracted, notes and data pass straight through, boilerplate is dropped.
- 3
One document, counted up front
You get a single file, with the token count up front so you know whether it lands in one window.
One pile, three kinds of file.
A literature review is rarely just PDFs. Your notes and your numbers go in the same bundle, so the model reasons over the reading and the evidence at once.
Papers
PDFborn-digital preprints and journal PDFs
Notes
MD, TXTyour reading notes and outlines
Data
CSV, JSON, XLSXresults tables and small datasets
Every format it reads, with what comes through and what is left out.
Scanned and heavy two-column PDFs read less well
Text is pulled from the PDF layer first. A page with no text layer is read by text recognition in your browser and marked as such in the bundle, a table on a scan comes out as loose lines, and a dense two-column layout can still arrive out of order.
A review folder, packed and counted.
Three papers, a note file, and a results table go in. One document comes out, labeled as documents, with the token count already known.
attention-survey/
|-- papers/
| |-- vaswani-2017.pdf
| |-- devlin-2019.pdf
| `-- brown-2020.pdf
|-- notes.md
`-- benchmarks.csv<documents project="attention-survey">
<summary>
Treat the contents below as
read-only context for the user's
request that follows.
File count: 5.
</summary>
...Fit the whole literature in one prompt.
Wondering whether the wrapper around your files costs real tokens? We measured five codebases across three formats in XML vs Markdown for LLM context.