nc://tabella

Tabella: read, convert and create office documents with pandoc. A Nodecode rewrite of dyiapanis/tabella by dyiapanis.

The tools are the original plugin's, with the same names and arguments:

Ask "turn this report into a PDF", "what does contract.docx say?", "give me the minutes as an EPUB" — the model reads with doc_read and writes with doc_convert or doc_create_document.

Get it

Tell the agent "add the tabella cell", or run nodecode add nc://tabella.

Set it up

The cell runs the programs you already have and needs two of them: pandoc (the converter; apt install pandoc, brew install pandoc) and pdftotext from poppler (apt install poppler-utils, brew install poppler). A PDF as output needs a LaTeX engine as well: xelatex or pdflatex (apt install texlive-xetex).

Then set the one folder documents are written into. In /settings, or in the tabella section of config.jsonc:

"tabella": {
  "output_dir": "~/Documents",   // required: where created and converted files go
  "pandoc": "pandoc",
  "pdftotext": "pdftotext",
  "pdf_engine": "",              // empty: xelatex if it is there, else pdflatex
  "max_chars": 200000
}

output_dir is the cell's one required member, and a start without it says so by name. Everything else has a default. A program named in the section may be a name on PATH or a path of its own.

Use it

You do not call the tools; the model does. A person can call the same verbs from eval, on a plain path, in the tabella package:

(tabella:convert-file "~/Documents/report.md" "pdf")
(tabella:create-file "docx" "## Agenda\n\n1. Budget\n2. Hiring" :title "Board meeting")
(tabella:read-text "~/Documents/notes.docx" :max-chars 4000)

(help :tabella) is the model's manual for the tools, kept while the cell runs. /tabella says where documents are written and which program is missing, and its panel reads or converts one.

What the guards are

pandoc can run code: --filter, --lua-filter and a format name ending in .lua all execute a file. The cell keeps the original plugin's two gates on that:

• Format names must be built-in pandoc names — letters, digits and underscores, with +ext/-ext modifiers — so nothing that looks like a path reaches -f or -t.
• extra_args takes only --toc, --number-sections and a one-token -Vkey=value. Anything else is refused, naming the flag.
• The category is a subfolder of output_dir; a category that climbs out of it (.., a leading slash) is refused.

A file written outside output_dir, when a call asks for one by name, is allowed and the answer says so: (outside ~/Documents).

What the rewrite leaves out

The original also built spreadsheets and slide decks (xlsx, xlsb, ods, odp, pptx) with python-docx, openpyxl, python-pptx and odfpy, wrote PDFs with WeasyPrint's HTML renderer, OCR'd images and scanned pages through anydoc and LiteParse, and redacted and re-inserted PDF text through PyMuPDF (doc_edit_pdf). This cell has none of those libraries: it writes documents with pandoc (markdown, docx, pdf, html, odt, epub, rst, latex), writes PDFs through LaTeX, and refuses an image or a scanned page by name instead of OCR'ing it. doc_read_meta answers what the filesystem knows rather than what a parser extracted. The plugin's MCP server, its loopback HTTP lane, its virtualenv bootstrap and its documents.convert/documents.create module layout are packaging for another harness, and are not here.

License

MIT. The tools, their schemas and their guards are dyiapanis's, rewritten in Common Lisp: see NOTICE and LICENSE.