TheThorn Tools

A text editor for prose. It opens in insert mode and behaves like an ordinary text box until you press Esc, at which point the whole vim vocabulary is there — counts, motions, text objects, .-repeat, visual mode, search. Commands are read as characters rather than key codes, which is why $ and ^ work on a keyboard that is not American.

Undo coalesces a run of typing into one step, so Ctrl+Z walks back by phrases and not by keystrokes. Whatever is on screen is written to a session file every few seconds and restored on next launch, so closing the window without saving does not throw the text away.

A small dragon sits on the page. It breathes, it blinks, its pupils drift toward the text while you type — and it never speaks unbidden. Clicked, it will tell you that your last sentence ran to fifty-one words, or that you have used merely three times in a paragraph. Those observations are computed from your document, locally, and nothing else reads it.

A cell can be an R expression. Not a macro, not an add-in, not a bridge you configure — a formula, sitting in a cell, recalculating when its inputs change, exactly as =SUM() does. =RDF("coef(lm(score ~ hours, df))", A1:B6) fits the model and spills the coefficients into the cells below. Python and Julia get the same treatment. If the expression draws, the picture is the cell.

Nothing is embedded: whichever of R, Python and Julia you already have is started as a child process speaking line-delimited JSON over a pipe, and it stays warm, so two hundred R formulae pay R's startup cost once. If none of the three are installed, everything else still works.

Seventy-five native functions handle the statistics you should not have to leave for, and three of them deliberately disagree with Excel: -2^2 is −4 here as it is in mathematics, SLOPE takes x first, and SKEW reports a population moment. Export to .xlsx translates through the parser rather than by search-and-replace, so the meaning survives even where the text cannot.

Nodes you write in, links between them, and a canvas that never rearranges your thinking between sessions. Position is truth. Ctrl+T tidies the subtree under the selection — but it is a command you invoke, not a constraint the canvas enforces, because auto-layout you cannot refuse turns a map into a diagram of the tool.

Drop a PDF on a node and it goes into the map. A .thmap is a ZIP whose first entry is the line-based, diffable map text and whose remaining entries are your documents, byte for byte. A link can rot; a document cannot. Both exist, deliberately.

Ctrl+P prints two documents into one PDF, and both are wanted: the plate, the canvas as resolution-independent vector art that prints as cleanly at A0 as at A5; and the outline, the same map as indented text with every body in full and an appendix naming each embedded document. A map you can only look at is a poster. A map you can only read is a list.

A reference manager where the PDFs live inside the library file. A .bib beside a folder of papers is two sources of truth that drift — a renamed file, a re-export that drops your notes, a synced folder that half-arrives. A .thshelf is one file you can mail, back up, or open in ten years, with the metadata and the papers together. BibTeX is an import and an export format, not the storage format.

The search box always has the keyboard. Just type, and it filters on every keystroke across author, title, venue, year, key, tags and your notes, because a library is not something you scroll — it is something you narrow. Drop a PDF on the window and it reads the XMP packet, then the Info dictionary, then hunts for a DOI through the inflated content streams, and tells you in the status bar which of the three it believed.

What is stored is not what is shown: the field keeps {\v S}m{\'\i}d so your exported .bib is byte-faithful, while the screen and the printed page say Šmíd — and file it under S, not under M.

A briar is a thorny wild rose. It is built from the other tools' bones — the same hand-drawn text engine, the same modal editing carried over whole from thorn edit — with a real R prompt underneath: a warm Rscript child, session state persisting across commands, Tab completion through R's own machinery, and ?lm doing what you expect.

Evaluation runs on its own thread, so a ten-minute fit never freezes the page, and Esc is a real interrupt — R's own interrupt condition, caught by the kernel, answered mid-flight, objects intact. Ctrl+Shift+` opens another R session in its own tab; a model fitting in one does not stop the other from answering.

Panels slide in one at a time and leave when you press Esc: environment, plots, help, files, packages — and a copilot that rides the AI subscription you already pay for, by shelling out to claude -p (or codex, or gemini), so no API key lives in any file. And View(df) does not open a grim grey grid: the frame is handed to thorn sheets, which knows what to do with one.