Not scanned —
made readable for Claude
Extract only the text from an HWP file and the tables fall apart, leaving numbers without their place. We process documents with merged cells, form fields, and outline levels intact, so Claude finds, reads, and edits on top of that structure.
Table 3 Activity budget · 7 rows × 4 cols · 2 merged areasExtracting text vs.
making a document understandable
In Korean official letters, budgets, and contracts, the tables and formatting are the content. So we hand over the structure.
| Scan or text extraction only | GCsoft processing | |
|---|---|---|
| Tables | Cells run into one line; merges disappear | Row/column coordinates and merge ranges preserved |
| Numbers | Claude guesses which item an amount belongs to | Amounts and dates extracted with position, quoted from the table |
| Form fields | Can't tell a blank from content | Read and filled by field name |
| Outline | □ ❍ - levels blur into plain text | Outline structure read; new documents follow the same levels |
| Location | "It was around here somewhere…" | Page, table number, and cell |
| Editing | People retype what Claude wrote | Original untouched; a new version with only changed cells |
| Finding | Needs the exact file name | Contents indexed; searches widen to similar terms and years |
One finds, one edits
The processing lives in two connectors. Use either alone, or both so "find it and fix it" happens in one request.
Find scattered documents and make them readable
Text from HWP, PDF, and Office files on the NAS is extracted and indexed in advance, so Claude finds them without opening each file.
- Search names and contents together, several phrasings at once
- Similar terms and years widened automatically; copies grouped
- Numeric tables read with merged cells preserved
- Scanned PDFs and photos checked directly as page images
- Each person's folder permissions applied
Read the structure, edit while keeping the format
Documents are read by outline, table grid, and form field. Edits leave the original alone and stack versions (v1, v2…).
- Outline, table coordinates, form fields, amounts and dates
- Locate content by page, table, and cell
- Edit cells, fill fields, replace text
- Draft official documents with cover, contents, levels, page numbers
- Compare versions, undo, preview pages
- 150+ Hangul features: merge cells, footnotes, redaction, PDF export
From original file to Claude
Originals stay put
Documents stay on your NAS or file server, never modified.
Content index
Text from HWP, PDF, and Office files is indexed for search. New files are rescanned and added.
Structure
Table grids, merge ranges, form fields, outline levels, and page positions in a form Claude can work with.
Claude works
Claude finds, reads, and edits on that structure. Results become new versions; people decide what to apply.
Anywhere Korean documents pile up
Public institutions · Schools
Letters, plans, and budgets in fixed HWP formats. Finding last year's version and updating it for this year.
Companies · Offices
Contracts, quotes, and reports on a NAS. Checking numbers across documents and making them match.
Individuals · Small businesses
Paperwork scattered on a home or office NAS. Asking Claude, "Where's that document?"
See the full tool lists in use cases and tools.
Common questions
Can't we just convert HWP to PDF and upload it to Claude?
The text survives, but merged cells, cell positions, and form fields are lost, so Claude has to guess which number belongs where — and people still have to copy any edits back into the HWP file.
Does it work with scanned documents?
Scanned PDFs and photos whose text can't be extracted are checked directly as page images.
Will the originals change?
No. Edits stack as versions (v1, v2…) on a copy, and people decide whether to apply them to the original.
Do documents have to leave our servers?
No. Documents stay on your NAS or file server; the connectors find and read them in place, with each person's folder permissions applied.
Ready to let Claude read your HWP files?
Tell us what documents you have and where. We'll settle the connection and processing scope together.